Supervisor: Jan Marton, Emmeli Runesson and Niuosha Samani Master Degree Project No. 2015:18 Graduate School

Master Degree Project in Accounting

Tone Management and Earnings Management

A UK evidence of abnormal tone in CEO letters and abnormal accruals

Sara Carlsson and Rania Lamti

Acknowledgements

We would like to take the opportunity to thank all the people who have contributed to the

accomplishment of this thesis and supported us during the past four months. The execution of

this thesis has been both interesting and challenging, and it is with joy that we present the

final product. First of all, we thank our supervisors, Jan Marton, Emmeli Runesson and

Niuosha Samani, for their guidance and valuable insights. Especially, many thanks to Emmeli

Runesson and Niuosha Samani who inspired us to study the subject of tone management. We

would also like to thank all of our seminar opponents, Carin Enocson, Veronica Rodriguez

Labra, Amanda Strandberg and Patricia Sandblom, for their encouragements and useful

advices.

Thank you!

Gothenburg, May 22nd

2015

Sara Carlsson Rania Lamti

Abstract Thesis: Master Degree Project in Accounting, School of Business, Economics and Law at the

University of Gothenburg, Graduate School, Spring 2015

Title: Tone Management and Earnings Management: A UK Evidence of Abnormal Tone in CEO

Letters and Abnormal Accruals

Authors: Sara Carlsson, Rania Lamti

Supervisors: Jan Marton, Emmeli Runesson, Niuosha Samani

Background and Problem Definition: The CEO letter is one significant narrative document through

which senior management have opportunity to express beliefs and values to their shareholders. The

CEO letter is unregulated in its nature and thereby subject to management opportunism through tone

management. Tone management could further be used to manipulate the perception of the firm, which

causes information asymmetry. Similar to the purpose of tone management, accruals could be

opportunistically managed in order to manipulate users’ perceptions of firm fundamentals. Thus,

managers could through CEO letters employ tone management to potentially hide earnings

management, and thereby mislead users about firm fundamentals.

Purpose: The purpose of this thesis is to test the possible association between tone management in

CEO letters and earnings management, using data from 2013 including firms listed on the London

Stock Exchange. Subsequently, the intention is to investigate if earnings management and tone

management are substitutes or complements. The purpose is additionally to test the relation between

tone management in CEO letters and financial performance

Research Design and Methodology: The theoretical framework of this thesis is used to develop

hypotheses, which subsequently guide the research forward. In the execution phase, a customized

dictionary is developed to fit with the purpose of the thesis. The execution phase continues in

correlation and multiple regression analysis. The outcome of the statistical tests contributes to fulfill

the purpose.

Empirical Results and Analysis: In accordance with previous literature, the empirical results reveal a

positive association between tone in CEO letters and financial performance. Strengthening the purpose

and the contribution of this thesis, empirical evidences reveal that abnormal tone in CEO letters and

abnormal accruals are positively associated.

Concluding Remarks: The tone in CEO letters is generally positive regardless of management

discretion. Furthermore the tone in CEO letters can be derived from firm performance, firm size and

annual stock return. Finally, the findings indicate that tone management and earnings management

through the use of abnormal tone and abnormal accruals functions as complements and are thus not

substitutes.

Key Words: Tone Management, Abnormal Tone, CEO letter, Earnings Management, Accruals

Table of Contents

1. Introduction .......................................................................................................................... 1 1.1 Background ....................................................................................................................................... 1 1.2 Problem Definition and Purpose ....................................................................................................... 1 1.3 Contribution....................................................................................................................................... 3 1.4 Delimitations ..................................................................................................................................... 3

2. Frame of Reference and Hypotheses Development ........................................................... 4 2.1 Tone Management ............................................................................................................................. 4

2.1.1 The CEO Letter .......................................................................................................................... 5 2.2 Earnings Management ....................................................................................................................... 6

2.2.1 Accruals-Based Earnings Management ..................................................................................... 7 2.3 Hypotheses Development .................................................................................................................. 8

3. Methodology ....................................................................................................................... 10 3.1 Research Design .............................................................................................................................. 10 3.2 Sample ............................................................................................................................................. 11 3.3 Variables .......................................................................................................................................... 13 3.4 Measuring Tone and Abnormal Accruals ........................................................................................ 15 3.4.1 Tone (TONE) ................................................................................................................................ 15

3.4.2 Abnormal Accruals (ABACC) .................................................................................................. 16 3.5 Statistical Analysis .......................................................................................................................... 17

3.5.1 Initial Data Analysis ................................................................................................................. 17 3.5.2 Correlation Analysis ................................................................................................................ 17 3.5.3 Multiple Regression Analysis .................................................................................................. 18

3.5.3.1 Regression models ........................................................................................................................... 19 3.6 Data Collection ................................................................................................................................ 19 3.7 Reliability and Validity ................................................................................................................... 19 3.8 Limitations....................................................................................................................................... 20

4. Empirical Results and Analysis ........................................................................................ 21 4.1 Descriptive Statistics ....................................................................................................................... 21 4.2 Correlation Analysis ........................................................................................................................ 22 4.3 Multiple Regression Analyses ......................................................................................................... 23

4.3.1 Hypothesis Testing ................................................................................................................... 24 4.3.1.1 Tone and Financial Performance ..................................................................................................... 24 4.3.1.2 Abnormal Positive Tone and Financial Performance ...................................................................... 25 4.3.1.3 Abnormal Tone and Abnormal Accruals ......................................................................................... 26

5. Discussion ............................................................................................................................ 28

6. Concluding Remarks .......................................................................................................... 31

7. Suggestions for Further Research .................................................................................... 32

List of References ................................................................................................................... 33

Appendix 1 .............................................................................................................................. 37

Appendix 2 .............................................................................................................................. 42

Appendix 3 .............................................................................................................................. 46

Appendix 4 .............................................................................................................................. 49

1

1. Introduction

The introducing chapter to this thesis aims to provide the reader with a background of the thesis subject.

Based on this, a discussion regarding the problem is presented and thereafter concluded with the purpose

of the thesis and hypotheses. Thereafter follows a statement of the contribution of this thesis to accounting

theory. Finally, the chapter terminates with a description of the delimitations surrounding the thesis.

1.1 Background

In the context of accounting, people often think about financial reporting as numbers. Interestingly, more

recent studies and analyses focus on accounting language as the medium through which companies

communicate to their externalities (Hales, Kuang & Venkataraman, 2011). Accounting language could be

expressed through narratives, which are becoming increasingly important as they become longer and

more sophisticated. Furthermore, narratives allow preparers of annual reports to disclose further detailed

information and explanations of events (Clatworthy & Jones, 2003; Merkl-Davies & Brennan, 2007). In

turn, the increased importance of accounting narratives helps reducing the information asymmetry that

could occur due to limitations in current accounting standards (Clatworthy & Jones, 2003; Huang, Teoh

& Zhang, 2014). Consequently, accounting numbers complemented with accounting narratives may better

indicate firm fundamentals.

The Chief Executive Officer (CEO) letter is one significant narrative document through which

management have opportunity to express beliefs and values to their shareholders (Amernic, Craig and

Tourish, 2010). The CEO letter is revealed by prior studies to be the most frequently read section in the

annual report (e.g. Courtis, 1998; Clatworthy & Jones, 2003). Thus, as the power of the CEO grows

(Amernic et al., 2010) and importance of accounting language becomes increasingly vital, the rhetorical

tone, i.e. the usage of positive and negative expressions, assessed in CEO letters is proved to be central

for the perception of the firm.

Through increased volume of accounting narratives, corporate leaders may or may not take advantage of

the opportunities to obfuscate perceptions using tone, i.e. tone management. In extension, Schipper

confirms another phenomenon of management manipulation namely, earnings management, in her

commentary paper published in 1989. Prior and recent studies on earnings management confirm, by

empirical evidence, its existence. One context of earnings management is accruals-based earnings

management, which appear to be present as the current accrual system of accounting allows for earnings

management behavior (Teoh, Welch and Wong, 1998).

1.2 Problem Definition and Purpose

Prior literature describe the CEO letter, or equivalent documents such as Management Discussion &

Analysis (MD&A) and chairman's letter, as an important communication part of the annual report, and

argues for its influence on investors and other users of the annual report (e.g. Abrahamson & Amir, 1996;

Amernic & Craig, 2007; Patelli & Pedrini, 2014). Through the CEO letter, management are allowed to

express beliefs, values and attitudes to shareholders and the letter thus provide insight to the motives and

2

intentions of CEOs (Amernic et al., 2010). The CEO letter could therefore provide an overview of firm

activities and performance (Clatworthy & Jones, 2003). The Financial Reporting Council (FRC), an

independent regulator in the UK, provides guidelines for the writing of annual reports that applies to all

parts, including the CEO letter (FRC, 2015). However, these guidelines are not mandatory (FRC, 2015),

and the content of CEO letters is consequently unregulated. Together with the relative freedom in

choosing the informational content and the lack of restrictions surrounding the presentation, the CEO

letter is an interesting document for analysis (Abrahamson & Amir, 1996).

Consequently, accounting narratives, such as the CEO letter, are proved to be subjects to management

manipulation and opportunism (Clatworthy & Jones, 2006). Furthermore, the potential for self-serving

behavior is enhanced by the fact that in the UK, for example, auditors review narratives to ensure that

they are materially consistent with financial statements, but other aspects, such as tone, are not considered

(Clatworthy & Jones, 2006). Clatworthy and Jones (2006) further explain that it might be natural for

individuals to “put our best foot forward”, which could explain opportunistic presentation of accounting

narratives. However, in the context of financial reporting, this thinking conflicts with basic financial

premises, such as true and fair view, if management deliberately misrepresents firm fundamentals

(Clatworthy & Jones, 2006).

According to Huang et al. (2014) tone management could be used as tool to obscure firm fundamentals.

The authors further define tone management as abnormal tone, i.e. a tone level inconsistent with firm

fundamentals. In addition, abnormal tone could further be practiced opportunistically to improve

understanding of the financial reports and as consequence mislead investors by the employment of

positive words to hide poor performance (Huang et al., 2014). This in turn creates a problematic situation

as the users of the annual reports hold less accurate perceptions about the firm. Preparers of annual reports

possess knowledge about the above discussed fact, that the CEO letter is the most frequently read section

(Courtis, 1998), and this awareness by itself reasonably creates incentives for preparers to manage tone in

order to change perceptions, without presenting false data. Thomas (1997) expresses her thoughts about

the problem surrounding managers’ letters by stating that “Managers’ letters suggest and imply, but they

do not lie” (p.63).

Hence, managers with incentives to manipulate users’ perceptions could select to present information in a

positive or negative manner. In connection, Huang et al. (2014) refer to prior studies on accruals-based

earnings management, suggesting that accruals, similar to the purpose of tone management, are managed

in order to manipulate users’ perceptions of firm fundamentals. Teoh et al. (1998) explain that earnings

consist of cash flows from operations and accruals. Accruals are adjusted by management in order reflect

future business transactions, that is, reflecting firm condition more accurately. However, Teoh et al.

(1998) emphasize that this flexibility allowed by the accounting system creates opportunities for earnings

management. Accruals that are managed opportunistically (hereafter abnormal accruals) are thus proxies

for earnings management.

To conclude, when firms engage in tone management, users of the financial reports may be required to

read between the lines. Managers could through CEO letters employ tone management to potentially hide

earnings management behavior, and thereby mislead users about firm fundamentals. Thus, the main

purpose of this thesis is to test if earnings management and tone management are substitutes or

3

complements. The purpose is additionally to test the relation between tone management in CEO letters

and financial performance. The main and the additional purpose are achieved through creating a

customized dictionary in order to measure tone, using data from 2013 including firms listed on the

London Stock exchange. Furthermore, the purposes are fulfilled through testing and analyzing the below

hypotheses1. Thereby, these hypotheses constitute the core of this thesis.

H1 - Tone in CEO letters is positively associated with financial performance

H2 - The probability of observing abnormal positive tone in CEO letters increases as financial

performance decreases

H3 - Abnormal tone in CEO letters is positively associated with abnormal accruals

1.3 Contribution

This thesis provides several contributions to accounting theory. Firstly, it provides empirical evidence and

in-depth analysis of an association between tone management and earnings management. Prior studies

concerning tone management primarily investigate the effect on investors and stock price reactions,

caused by managerial opportunism (e.g. Feldman, Govindaraj, Livnat & Segal, 2010; Tan et al., 2014).

Furthermore, previous scholars include accruals as control variable when studying tone (e.g. Feldman et

al., 2010; Li, 2010; Huang et al., 2014). Although Huang et al. (2014) present a correlation between tone

management and accruals, the main purpose of their study, and previously mentioned studies, is not to

analyze the relation between tone management and earnings management. Furthermore, these studies test

the existence of tone management primarily in MD&As and earnings press releases. Thus, the

combination of tone management in CEO letters and earnings management is, to our knowledge,

unexplored.

In addition, the general focus on previously conducted tone studies has been listed or unlisted firms in the

US. On the contrary, this thesis approaches firms listed on the London Stock Exchange, a large and liquid

stock market characterized with dispersed ownership. Finally, the dictionary created for the purpose of

this thesis is customized to fit CEO letters and could therefore apply to future tone studies.

1.4 Delimitations

With the limited time frame in mind, this master thesis covers a limited scope of the research field.

Firstly, regarding tone management, the document to study is the CEO letter of annual reports. Secondly,

in terms of textual analysis, neither readability in the shape of reading ease, nor graphic presentations, are

considered when measuring tone. In the context of earnings management, only accruals-based

measurements are applied.

1 Hypotheses are developed in section 2.3

4

2. Frame of Reference and Hypotheses Development The following chapter constitutes the theoretical framework used in this thesis and aims at summarizing

relevant arguments provided by existing literature within the chosen fields i.e. tone management and

earnings management. Finally, hypotheses for this thesis are derived from the literature and developed in

section 2.3.

2.1 Tone Management

The linguistic features and textual analysis of accounting have gained attention during recent years. Li

(2008) examines reading complexity in annual reports through measuring annual report readability in

MD&As and in notes to financial statements. The findings indicate that firms with lower and less

persistent earnings report disclosures that are difficult to read, in other words, disclosures with low

readability. Interestingly, the author suggests a clear association between linguistic features in reported

disclosures and firm performance, concluding that firm performance is positively correlated with annual

report readability. Additionally, Li (2008) continues beyond readability by exploring other lexical features

in disclosures, such as the use of positive versus negative words (tone), revealing that loss firms that use

positive words to a larger extent than negative words in their MD&As have less persistent earnings. In

accordance, Rutherford (2005) explains that stylistic choices such as lexical features, words choices,

frequency use and word complexity affect the perception accounting narratives. More so, Rutherford

(2005) concludes that management’s letter to shareholders in general is positively charged, regardless of

financial result.

The notion of tone management has been the subject of several recent papers, both in the context of

accounting and in the context of other research fields. Within accounting and financial reporting, scholars

use the concept mainly in order to establish how tone management would or could influence investors

(e.g. Davis & Tama-Sweet, 2012; Huang et al., 2014; Tan et al., 2014). Huang et al. (2014) further define

tone management as: “... the choice of the tone level in qualitative texts that is incommensurate with

concurrent quantitative information…” (p.1083). According to Huang et al. (2014), the rhetorical use of

narratives is important in order to understand quantitative information. However, when agency motives

are present, the rhetoric could instead mislead the reader and thereby be used strategically rather than

informative (Huang et al., 2014). Thereof, Huang et al. (2014) investigate whether tone management is

used for strategic purposes or for informative purposes, and if the capital market discounts for strategic

motives when reacting to earnings announcements. When firm fundamentals are better than indicated by

quantitative information, due to limitations in accounting standards, tone management could be employed

for informative purposes (Huang et al., 2014). However, tone management may be used for strategic

purposes to change perceptions about firm fundamentals (Huang et al., 2014). Extending the research by

Li (2008), Huang et al. (2014) are of the opinion that tone is jointly affected by firm fundamentals and

managerial incentives to report strategically. They find that despite opportunities for truthful disclosures,

management abuses abnormal positive tone when firm fundamentals are poor. As consequence, the

authors find abnormal positive tone to be negatively correlated with future earnings. Additional findings

reveal that abnormal positive tone is associated with positive immediate market reactions to earnings

5

announcements, followed by negative market reaction in subsequent periods. The authors conclude this

reversal effect to be driven by the overestimated reaction to abnormally positive earnings announcements.

Tan et al. (2014) refer to Huang et al. (2014) explaining that language sentiment effect is associated with

the use of positive versus negative words. By holding the information content constant, Tan et al. (2014)

investigate whether tone management is used to influence investor perceptions. In contrast to remaining

studies on tone management, Tan et al. (2014) explore the co-occurrence of language sentiment and

readability in earnings press releases. In an additional analysis, the authors explain language sentiment to

be an effect of attribute framing, which is the situation when people’s perceptions about identical items

differ depending on the extent to which the items are described in a positive or negative manner. Drawing

on attribute framing, the authors find that positive language (as opposed to neutral language) leads to

positive framing effect as the participants of the study view firm performance to be better than indicated

by financial figures, regardless of investor sophistication. However when taking the co-occurrence of

language sentiment and readability into account, the authors find that when readability is high, the

language used becomes less significant, regardless of sophistication level. In contrast, when readability is

low, language sentiment is proved to affect investor judgment depending on sophistication level (Tan et

al., 2014). Less sophisticated investors are more influenced by positive framing (positive tone) despite the

fact that the information provided in annual reports may be inconsistent with firm fundamentals, leading

investors to make overestimations of earnings (Tan et al., 2014). In contrast, the authors emphasize, when

readability is low, more sophisticated investors regard positive tone as less credible and thereby make

lower earnings judgments than firm fundamentals. Tan et al. (2014) conclude that underestimated

earnings cause the use of positive language to backfire.

The common feature behind the reviewed literature above is that management incentives behind the use

of language appear to determine the tone. According to Courtis (1998), firms subject to media attention

might have motives to influence readers’ perceptions and engage in tone management in order to restrict

outside interference. In accordance, Huang et al. (2014) identify that tone management is used when there

are incentives to mislead investors, for example when management have targets to meet or beat. This

argument is further strengthened by Henry (2008) explaining that firm’s ability to manage tone is

dependent on firm fundamentals compared to analyst forecast. Under these circumstances, she explains

that tone management could be achieved by describing outcomes and events with a positive language and

provides positive comments about the future. Furthermore, Davis and Tama-Sweet (2012) emphasize that

management have incentives to manage tone in earnings press releases to minimize stock price effects on

negative news by applying less pessimistic language. This reasoning by Davis and Tama-Sweet (2012) is

in accordance with both Huang et al. (2014) and Tan et al. (2014), who argue that negative financial

performance may be disguised using positive tone.

2.1.1 The CEO Letter

The reviewed studies in above sections use different channels of management communication when

assessing tone management. Several scholars (e.g. Li, 2010; Davis & Tama-Sweet, 2012) apply the

MD&A report when determining the implications of tone management, because of the voluntary nature of

its content and excessive room for management discretion. Other scholars (e.g. Huang et al., 2014; Henry,

2008; Davis & Tama-Sweet, 2012), analyzing tone management effect on stock prices, target earnings

press releases as source of management communication. Earnings press releases are widely used due to

6

investors’ timely reactions, creating incentives for management to act strategic in their use of language

(Davis & Tama-Sweet, 2012).

Other authors (e.g. Hildebrandt & Snyder, 1981; Amernic & Craig, 2007; Amernic et al., 2010; Patelli &

Pedrini, 2014) rely on the annual letter to shareholders (hereafter CEO letter) when studying tone

management. In general, the CEO letter includes statements summarizing past events and future prospects

(McConnell, Haslem & Gibson, 1986). The CEO letter is neither audited, nor are there specific

requirements regarding its content and shape (Fisher & Hu, 1988). Management is thereby free to provide

statements of anything it considers important (McConnell et al., 1986). Thereof, CEO letters contain other

information than provided in the financial statements, along with explanations and interpretations

(Abrahamson & Amir, 1996).

As presented in section 1.1, the CEO letter is claimed by several authors to be the main communication

channel for management to review firm performance to shareholders. Amernic et al. (2010) claim “CEO

letters to shareholders in annual reports are important instances of the use of language in the disclosure of

senior corporate leaders. Such letters are narrative accountability texts offering valuable insights to the

motives, attitudes and mental models of management” (p.26). In accordance, Abrahamson and Amir

(1996) argue for the importance of CEO letters, implying that CEO letters include useful information in

making investment decisions. McConnell et al. (1986) state that CEO letters should not be disregarded

exclusively as the users of the documents otherwise could lose important signals. However, the authors

together with Fisher and Hu (1988) dissuade analysts and investors to solely rely on CEO letters when

making earnings forecasts or investment decisions. More so, McConnell et al. (1986) encourage investors

to read between the lines and take a “hard look” at the language of the CEO letter, since the language

could differ dependent on the financial performance of the firm. Additionally, McConnell et al. (1986)

present a relation between the assessment of prospective performance in CEO letters, and the firm’s

actual performance. Concluding, these results point to the usefulness of the CEO letter since it could be

used to assess future performance (McConnell et al., 1986).

2.2 Earnings Management

Earnings management is a notion that has been quite popular among researchers to explore (Healy &

Wahlen, 1999). One particular study that has been cited by several scholars (e.g. Leuz, Nanda &

Wysocki, 2003; Burgstahler, Hail & Leuz, 2006) is the literature review on earnings management by

Healy and Wahlen, conducted in 1999. Healy and Wahlen (1999) define earnings management as the

situation when management abuses the opportunity of judgment, either to mislead stakeholders about firm

fundamentals or to influence contractual outcomes. Accounting standards are constructed to fit different

accounting environments, and the element of judgment is therefore necessary (Healy & Wahlen, 1999).

However, this requires users to make accounting decisions based on privately held information, which

potentially creates discretionary reporting situations and agency problems (Burgstahler et al., 2006).

Burgstahler et al. (2006) further explain that management, consequently, either can construct earnings as

less or more informative dependent on the usage of privately held information.

In addition, the desire to mislead stakeholders could according to Healy and Wahlen (1999) include

limiting stakeholders’ access of information in order to reduce transparency of information. In

7

accordance, Leuz et al. (2003) argue that firms have incentives to mislead stakeholders by misrepresent

firm performance, thereby mask true performance through earnings management as a result of conflicts

between firms and their stakeholders. Similarly, Schipper (1989) possess a view of accounting numbers as

information and relates earnings management to “disclosure management”, emphasizing that managers

intervene financial reporting to obtain private gain. Schipper (1989) further argues that earnings

management could occur everywhere in external reporting and undertake all shapes.

The incentives to manipulate earnings are widely explored in previous literature. Dechow and Skinner

(2000) emphasize that capital market incentives for earnings management is the most accurate focus since

stock market prices and their relation to earnings has become increasingly important. In turn, managers

have incentives to manage earnings to both maintain and improve stock price valuation, which is

supported in the review by Healy and Wahlen (1999). While these studies connect the capital market to

earnings management, Leuz et al. (2003) link how incentives to manage earning could be associated with

institutional factors. Leuz et al. (2003) investigate how the level of earnings management could vary

across clusters and find that investor protection rights is a key determinant for earnings management.

Additionally, Dechow et al. (2010) indicate that when firms have targets to meet or beat, set by them or

by outside parties, they also have incentives to manage earnings.

2.2.1 Accruals-Based Earnings Management

Earnings consist of total accruals and cash flow from operations, and the amount of accruals therefore

affect the amount of reported earnings (Teoh et al., 1998). Teoh et al. (1998) explain that by upwardly

adjust accruals today; managers can increase current reported earnings at the expense of future reported

earnings. However, Richardson (2003) informs that high levels of accruals could be unintentional due to

the accounting environment of the firm, and thereby not a result of earnings management. In fact, accruals

are the core of the modern accounting system and are used to prevent mismatches between short-term

transactions (Runesson, 2014). Nevertheless, the nature of accruals require high amount of estimation of

future events and subjective allocation of past transactions, and are therefore subject to earnings

management behavior (Richardson, 2003).

Due to the role of accruals within the accounting system, scholars generally decompose accruals into two

separate components, normal (non-discretionary) and abnormal (discretionary). Normal accruals are those

related to the fundamental performance of the firm, and, abnormal accruals are in contrast those

exceeding the normal (Dechow et al., 2010). Hence, abnormal accruals are not explained by firm

fundamentals and can therefore function as indicator of management’s abuse of reporting flexibility

(Healy & Wahlen, 1999; Geiger & North, 2011). In accordance, both prior and recent scholars (e.g.

Godfrey et al., 2003; Li 2010; Huang et al., 2014) argue that management uses abnormal accruals as tools

to manipulate investor perception about true firm performance. Hence, research indicates that abnormal

accruals are used to manage earnings, and abnormal accruals are therefore considered an appropriate

proxy for earnings management among researchers.

Subsequently, the effect of accruals has been widely explored within earnings management literature.

Previously mentioned Teoh et al. (1998) examine whether income-increasing accruals accounting leads to

overly optimistic investor judgments of stock issues. The authors both predict and conclude that

overvaluation of performance leads to poor earnings and stock return performance in subsequent periods.

8

By decomposing earnings into accruals and cash flow from operations, the authors find that the

overvaluation and the subsequent earnings underperformance are caused by accruals. These findings are

in accordance with Sloan (1996), who finds that earnings are less persistent when highly dominated by

accruals. After further decomposing the accrual component into abnormal and normal, Teoh et al. (1998)

provide evidence consistent with earnings management that abnormal accruals both predict

underperformance of post-issue earnings and post-issue stock returns. Additional evidence of the

connection between low persistence of earnings and earnings management is provided by Dechow, Sloan

and Sweeney (1995). The authors present empirical evidence of the reversal effect of accruals, i.e.

increase (decrease) in year zero followed by decrease (increase) thereafter. According to Sloan (1996), the

reversal effect is a sign of earnings management contributing to the lower persistence of the accrual

component of earnings.

The low persistence of earnings dominated by accruals has more recently been explored by Dechow et al.

(2010). The authors examine the notion of earnings quality and present a review of its proxies,

determinants and consequences. High quality earnings are defined as earnings highly informative about

the firm’s financial performance (Dechow et al., 2010). Thereby, proxies for earnings quality are

transactions that might lower the informational use of earnings. One such category of proxies are

according to Dechow et al. (2010) the property of earnings, including among others earnings persistence,

accruals and target beating. Furthermore, the authors mention that the majority of studies published on the

subject, view abnormal accruals as the determinant of earnings quality since abnormal accruals are

assumed to erode decision usefulness. Thereby, accruals-based earnings management is assumed to erode

earnings quality since manipulated earnings have low persistence (Dechow et al., 2010).

2.3 Hypotheses Development

Throughout the literature supporting this thesis, researchers have highlighted the relationship between

firm performance and tone in various types of management letters. For example, Davis and Tama-Sweet

(2012) test the use of language in financial reports where they expect current firm performance to be the

significant determinant of the use of positive versus negative language. In connection, authors (e.g.

Amernic & Craig, 2007; Merkl-Davies & Brennan, 2007; Henry, 2008) express that tone is managed

depending on whether companies face high or low profits, arguing that when profits are high,

management tend to use a positive tone in accounting narratives as a result of good management. Huang

et al. (2014) state that when presentation of information is neutral, the optimism in tone is positively

correlated with performance. This reasoning is in accordance with Davis and Tama-Sweet (2012), who

provide similar evidence regarding optimistic tone and financial performance. Thereof, the common

thought among previously mentioned scholars appears to be that the tone in CEO letters is generally

positive, and that the association between tone and financial performance consequently is positive.

However, the above noted studies present empirical results on firms listed in the US whilst investigating

various management communication channels. Thus, the thought of a similar association between tone in

CEO letters and financial performance of firms listed on the London Stock Exchange is thereby justified,

and presented as H1.

H1 - Tone in CEO letters is positively associated with financial performance

9

In contrast to above, Hildebrandt and Snyder (1981) conclude that positive words are more frequently

present in CEO letters than negative words, regardless of financial results. The authors explain that a

financially bad year will include positive statements that do not reflect firm fundamentals, i.e. abnormal

usage of tone. Similarly, Rutherford (2005) presents results indicating that management letters are

dominated by positive words, which justifies the reasoning by Hildebrandt and Snyder (1981).

Furthermore, Rutherford (2005) argues that loss making firms employ word such as profit and profits,

more frequently than profit making firms. The research by Hildebrandt and Snyder (1981) and Rutherford

(2005) legitimate the thought of potential usage of abnormal positive tone when firm performance is

negative. Drawing on Huang et al. (2014), one can assume that loss making firms have incentives to

change users’ perceptions, and thereby more prone to employ abnormal positive tone in CEO letters

compared to profit making firms. The following hypothesis is thereof presented:

H2 - The probability of observing abnormal positive tone in CEO letters increases as financial

performance decreases

Referring to previously presented descriptions of tone management and earnings management, mutual for

both management behaviors is the intent to affect or shape perceptions about financial performance.

Adding the resembling incentives behind the two behaviors, the thought of coexistence is justified. Li

(2010) explains that both accruals and tone could by managers signal future firm performance. In

addition, when incentives to mislead investors are present, there may be a positive association between

accruals and the tone of MD&As (Li, 2010). Similarly, Huang et al. (2014) investigate, in additional

analysis, the co-occurrence of tone management and earnings management to provide evidence on

whether the two behaviors complement or substitute each other when manipulating investor perceptions

through earnings press releases. In sum, the authors found that the association between abnormal accruals

and abnormal tone is statistically significant, proving management's’ usage of both behaviors

simultaneously in earnings press releases. These evidence are also provided by Schrand and Walther

(2000) stating “strategic disclosure in earnings announcements is related to earnings management…”

(p.3). Hence, one can predict a similar association between abnormal tone in CEO letters and abnormal

accruals of firms listed in UK. The third and last hypothesis is thereby proposed as:

H3 - Abnormal tone in CEO letters is positively associated with abnormal accruals

10

3. Methodology The aim of this chapter is firstly to present the research design, which is illustrated in figure 3.1.1 and

aims to serve as guidance throughout the thesis. Thereafter follows a description of the sampling process.

Subsequently, the chosen variables are presented in table 3.3.1 and described one by one. Thereafter,

models regarding measurements of the variables tone and accruals are of particular importance for this

thesis and therefore presented in detail in section 3.4. Afterwards, the selected statistical analyses for this

thesis are presented and followed by the data collection. Finally, comments about validity, reliability and

limitations are provided in the last subsections.

3.1 Research Design

Quantitative research methods are perceived as most suitable when testing the developed hypotheses.

Thereby, aligned with positivism, the research is based on a theoretical structure, which subsequently is

tested through empirical observations. In other words, the logic of the research could be identified as

moving from the general to the specific (Collis & Hussey, 2014). The magnitude of the variables is tested

through the use of hypotheses, which guide the research forwards. Thus, the outcome of the statistical

tests will through analysis contribute to fulfill the purpose of the thesis. The hypotheses are constructed

through in-depth review of existing literature, and thereby test theoretical propositions against empirical

evidence (Collis & Hussey, 2014). The hypotheses, developed in section 2.3, are presented below:

H1 - Tone in CEO letters is positively associated with financial performance

H2 - The probability of observing abnormal positive tone in CEO letters increases as financial

performance decreases

H3 - Abnormal tone in CEO letters is positively associated with abnormal accruals

The below figure (3.1.1) functions as guidance towards the main purpose of the thesis, which is stated as

H3. As illustrated, tone and accruals, which are measured separately, both constitute of a normal and an

abnormal component. The abnormal components, calculated as regression residuals, are subsequently the

main elements when testing H3, which explores a potential association between abnormal tone (ABTONE)

and abnormal accruals (ABACC). Abnormal tone and abnormal accruals thereby functions as proxies for

discretionary management behavior.

11

Figure 3.1.1 Research Design

H1 and H2 explore the association between tone and firm performance, and thus contribute to the

additional purpose of this thesis. H1 establish the association between tone and financial performance

regardless of management discretion, which is of analytical importance. Furthermore, the usage of

positive tone when performance is negative, i.e. abnormal positive tone, is of particular interest for this

thesis. H2 aims to establish the probability of observing abnormal positive tone as financial performance

decreases and is therefore expressed in logistic terms.

3.2 Sample

The population constitutes of firms traded on the London Stock Exchange during year 2013. The UK is

according to Leuz et al. (2003) classified as large in terms of stock market, and characterized by low

ownership concentration. Thereby, the information provided by firms listed on the London Stock

Exchange is of importance and followed by great deal of analysts and investors.

Following Geiger and North (2011), financial services, i.e. banks and insurance firms, are excluded from

the population. The accruals of banks and insurance firms are considered industry specific and would, if

TONE ACCRUALS

NORMAL ABNORMAL (ε)

ABNORMAL (ε) NORMAL

TONE MANAGEMENT EARNINGS

MANAGEMENT

H3

12

included, disturb the measurement of accruals (Stubben, 2010). The firms are identified through using

two-digit Industry Classification Benchmark (ICB) code2 and thereafter deleted from the population.

A prerequisite of this thesis is that the required data exist in the Datastream database. Therefore, in order

to preserve the sample consistent, observations with missing values related to the dependent, independent

and control variables are eliminated. In order to limit the risk of potential comparison issues, the currency

of the variables are, through functions of Datastream, converted into GBP (£) before choosing a sample.

Moreover, firms with no available annual report or no CEO letter or equivalent letter are excluded. Based

on the mentioned restrictions, the final sample consists of 415 firms (see Appendix1). The sample size is

believed to be suitable since the authors execute the content analysis of the CEO letters manually. The

sampling process is visualized below.

Table 3.2.1 The Sampling Process

Initial number of firms provided by ESMA 1410

Firms categorized as financial services - 34

Deleted due to missing values - 827

Deleted due to annual report not found - 36

Deleted due to no CEO letter or equivalent - 98

FINAL SAMPLE 415

2 Firms classified with two-digit ICB code, 8300, 8500 or 8700.

13

3.3 Variables Table 3.3.1 Summary of Variables

Name Abbreviation Type of Variable Proxy for Measurement

Tone TONE Dependent Level of Tone

(Positive words-negative

words)/(positive

words+negative words)

Abnormal Tone ABTONE Dependent Tone

Management Regression residual

3

Abnormal Positive

Tone ABPOS Dependent

Tone

Management

1=Abnormal tone >0

0=Abnormal tone ≤0

Abnormal Accruals ABACC Independent Earnings

Management Regression residual

Performance LOSS Independent/Dummy Financial

Performance

1=EARN<0

0=EARN≥0

Profitability EARN Independent/Control Financial

Performance Net income/total assets

Size SIZE Control External

Attention

Log(market value of

equity)

Book-to-Market

Ratio BTM Control

Growth

Potential

Book value of

equity/market value of

equity

Annual Stock

Return RET Control

Growth

Potential ((P1-P0)+Div)/P1

4

Tone, TONE

Drawing on the research by Huang et al. (2014), the firm’s total level of tone is interpreted as the normal

level of tone together with the abnormal level (ABTONE). Normal level of tone symbolizes the neutral

description of firm fundamentals, and is expressed through the control variables of regression (1) (see

section 3.5.3.1). Accordingly, the TONE variable expresses the tone level including both the normal level

and the abnormal level. TONE is used in H1 to calculate ABTONE and to establish the association

between tone and financial performance. The procedure of measuring TONE is further described in

section 3.4.1.

Abnormal Tone, ABTONE, and Abnormal Positive Tone, ABPOS

ABTONE is considered the dependent variable in regression (3). Following Huang et al. (2014) ABTONE

is calculated as the residual value when deducting the normal level of tone from the total level of tone,

and thereby proxies for the strategic usage of tone, i.e. tone management. If tone is expected to be

consistent throughout the years, the change in tone could also proxy for tone management (Huang, et al.,

2014). However, as this thesis only considers one firm year (2013), tone management is better indicated

3 Regression residual from regression (1)

4 P0=Stock price year-end, 2012, P1=Stock price year-end, 2013, Div=Dividends per share 2013.

14

by ABTONE. In contrast to Huang et al. (2014), ABTONE is considered to be both negative and positive

abnormal tone. The value of ABTONE is expressed through the error term in regression (1) (see section

3.5.3.1), and thereby calculated as the residual value.

In order to test the probability stated in H2, the ABPOS variable is created as a dummy variable set to 1 (0)

if ABTONE is positive (negative). The intention is to isolate the firms with positive abnormal tone in

order to test the probability of observing ABPOS (i.e. ABPOS = 1) as financial performance decreases.

Since ABPOS is derived from ABTONE, ABPOS also functions as proxy for tone management.

Abnormal Accruals, ABACC

Following previous literature, abnormal accruals (ABACC) are used as proxy for earnings management.

The variable is considered independent due to its use in H3, and is identified as the error term in the

Cross-Sectional Modified Jones Model (see section 3.4.2).

Financial Performance, EARN and LOSS

EARN and LOSS capture current financial performance, which according to the literature (e.g. Davis and

Tama-Sweet, 2012; Huang et al., 2014) is considered to affect the level of tone. The EARN variable is

calculated as the net income divided by total assets, also known as the Return on Assets (ROA). LOSS is a

dummy variable set to 1 when EARN is negative and 0 when EARN is equal to zero or positive. Following

Huang et al. (2014) EARN intents to measure profitability, whilst LOSS act as a performance benchmark.

EARN is included in regressions (1)-(3), and its correlation with TONE is tested in H1. LOSS is included

in regressions (2) - (3) and serves as the main independent variable along with EARN in H2.

As the value of LOSS is dependent on EARN, an interaction effect arises (Wooldridge, 2013). The

interaction variable, EARN_LOSS is the product of multiplying EARN with LOSS, and is used in

regressions (2) and (3) in order to control for the interaction effect.

SIZE

Following prior literature on tone management, size is generally controlled for as the external attention

drawn to the firm due to its size could affect the level of tone. For instance, Li (2010) bases his reasoning

on previous studies stating that larger firms may be more cautious in their expressions to avoid political

and legal costs. Moreover, Courtis (1998) suggests that firms subject to media attention might have

motives to influence readers’ perceptions, which is further reasoning for including size as a control

variable. Following Huang et al. (2014), SIZE is calculated as the logarithm of market value of equity

year 2013.

Future Growth Opportunities, BTM and RET

In accordance with Li (2010) and Huang et al. (2014), the Book-to-Market ratio (BTM) is controlled for

as it represents investment opportunities and growth potential, which could affect the level of tone.

Accordingly, Davis and Tama-Sweet (2012) expect that managers of high-growth firms have incentives to

report information strategically, which might affect the tone level. BTM is therefore considered an

appropriate control variable when investigating tone, and included in all tone regressions.

15

Similar to BTM, RET is also considered to capture future performance opportunities (Huang et al., 2014),

and is included in previous tone studies as a control variable. The RET variable is therefore included as a

control variable in all tone regressions. Hence, both BTM and RET are calculated based on the current

financial position. Nevertheless, the ratios include information about future performance beyond what is

conveyed in EARN, and thus represent future growth opportunities (Huang et al., 2014).

3.4 Measuring Tone and Abnormal Accruals

3.4.1 Tone (TONE) Scholars have used different approaches in order to measure the tone of various management documents;

however, there are two general approaches for conducting content analysis: the dictionary approach and

the statistical approach (Li, 2010). Out of these two, the dictionary approach appears more commonly

within financial research and “maps” words into different categories based on predefined rules (Li, 2010;

Loughran & McDonald, 2011). The statistical approach, on the other hand, relies on statistical techniques

such as measuring the correlation between the frequencies of certain key words (Li, 2010).

For the purpose of this thesis, the dictionary approach is considered appropriate, mainly as it is the most

commonly used approach within textual analysis (Li, 2010). Furthermore, the purpose is to measure

percentages of words within specific categories, i.e. positive and negative, which justifies the dictionary

approach (Li, 2010). The tone of CEO letters is thus measured by the amount of positive and negative

words included in the text by using a predetermined word list, i.e. dictionary. Within the dictionary

approach, the Harvard Psych sociological Dictionary (General Inquirer, GI) and the software program

DICTION are frequently used as categorization tools to evaluate the tone of financial texts (e.g. Henry,

2008; Loughran & McDonald, 2011; Craig & Brennan, 2012). However, Loughran and McDonald (2011)

found that particularly the GI dictionary is not designed with the purpose to fit financial research since

several words identified as negative (73.8%), typically are not considered negative in financial contexts.

The authors therefore address this issue and develop a word list, based on GI, more suitable to financial

research. Although the dictionary has been used in previous research, it is not considered appropriate for

the purpose of this thesis due to its magnitude. Henry (2008) studies the tone of earnings press releases

and its impact on investors, and provides a more manageable dictionary. Thereof, the dictionary provided

by Henry (2008) is used and serves as the foundation for the development of the customized dictionary

used in this thesis (words marked with asterisk in Appendix 2).

The dictionary of this thesis is created by manually reviewing CEO letters of various firms listed on the

London Stock Exchange. The letters are manually analyzed by their content in order to detect and classify

words as either positive (e.g. delighted, pleased, excellent) or negative (e.g. disappointed, unfavorable,

weak). In order to avoid bias, these letters do not refer to the firms included in the sample of this thesis.

The advantage with manual content analysis is that the content analysis becomes more precise, detailed

and tailored with regards to the specific research setting (Li, 2010). The final dictionary is presented in

Appendix 2. In line with Loughran and McDonald (2011) and Huang et al. (2014), if negations (no, not,

none, neither, never and nobody) are used immediately before a positive word, the positive word is

counted as negative. To the extent possible, grammatical and linguistic features are taken into

consideration with the purpose to capture the style of writing.

16

Due to access limitations, computer software such as DICTION is not used in this thesis. Instead, the

execution phase is conducted manually using a spreadsheet software searching for predetermined words.

Subsequently, the CEO letters of the sample firms are imported into the spreadsheet software and scanned

for words based on the customized dictionary. The tone of the text (TONE) is thereafter measured as a

frequency count of the words included in the dictionary. The TONE variable is calculated based on the

equation provided by Henry (2008):

𝑇𝑂𝑁𝐸 = (#𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 − #𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠) / (#𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 + #𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠) (𝑎)

The above equation equals to a scale within a minimum value of -1 and a maximum value of +1. TONE

equal to 0 suggests a neutral usage of positive and negative words whereas -1 (+1) suggests no usage of

positive (negative) words. The result from the above equation is thereafter used in the regression model to

identify the abnormal level of tone (see 3.5.2.1).

3.4.2 Abnormal Accruals (ABACC)

In order to measure the magnitude of accruals-based earnings management, a calculation of total accruals

is firstly executed, followed by a model to measure abnormal accruals (Healy & Wahlen, 1999). Total

accruals are subsequently regressed in order to identify which accruals that belong to the operating

activities of the firm e.g. sales revenue, accounts receivables and fixed assets (Healy & Wahlen, 1999).

Remaining accruals left undefined thus exceed the normal and could indicate earnings management.

Based on this along with prior studies on accruals-based accounting, the chosen models to measure

abnormal accruals for this thesis are presented below. Following previous studies, the accruals regression

is run for each one-digit ICB code combination in order to control for industry differences.

𝑇𝐴𝑐𝑐𝑗𝑡 = 𝐸𝐵𝐸𝐼𝑗𝑡 − 𝐶𝐹𝑂𝑗𝑡 (𝑏)

Where:

TAcc = Total accruals scaled by lagged total assets,

EBEI = Earnings before extraordinary items,

CFO = Cash flow from operations,

j= firm; and

t= year

𝑇𝐴𝑐𝑐𝑗𝑡 = 𝛽0 (1

𝐴𝑠𝑠𝑒𝑡𝑠𝑗,𝑡−1

) + β1(∆Sales𝑗𝑡 − ∆AR𝑗𝑡) + β2𝑃𝑃𝐸𝑗𝑡 + ε𝑗𝑡 (c)

Where:

Assets = Total assets,

ΔSales= Change in sales scaled by lagged total assets,

ΔAR= Change in accounts receivable from operating activities scaled by lagged total assets; and

PPE= Gross property, plant and equipment scaled by lagged total assets

ε= Regression residual5

5 Abnormal Accruals (ABACC)

17

Both models are extracted from Huang et al. (2014). The cross-sectional modified Jones model (c) is

chosen to measure abnormal accruals, firstly, as it is commonly used within accruals-based accounting

studies, and secondly, as the model takes credit sales manipulation into account (Teoh et al., 1998; Geiger

& North, 2011; Huang et al., 2014).

Following Huang et al. (2014), a constant (1/Assets) is included in the model. The inclusion of the

constant varies among scholars. However, Kothari, Leone and Wasley (2005) state the constant to be

important for the model, as the exclusion of it would bias the result and make the accruals measure less

symmetrical. The inclusion of the constant would thus enhance the ability to address the research problem

(Kothari et al., 2005). Additionally, abnormal accruals are expressed through the error term (ε). The

advantage with measuring accruals as the residual value (ε) is, according to Dechow et al. (2010), that

such models attempt to isolate the managed or error component of accruals, and the models have become

accepted within financial research. Thus, the identification of abnormal accruals is simplified.

3.5 Statistical Analysis

3.5.1 Initial Data Analysis An initial analysis of the variables is performed before executing further statistical analyses in order to

identify potential outliers, or extreme values, that might bias subsequent analyses. Variables that show

extreme skewness and kurtosis diverge from the normal distribution (Collis & Hussey, 2014) and are thus

winsorized to suit future analyses. The essence of winsorization is not excluding extreme values, but

rather alters the original data by setting extreme values equal to specified percentiles of the distribution

(Leone, Minutti-Meza & Wasley, 2014). In this thesis, extreme values are winsorized to the 1 and/or 99

percentiles. Although winsorization of data is common among researchers, Leone et al. (2014) emphasize

the potential problems with winsorizing data. The extreme values might not be results from error

computations, but could rather be effects of the business environment and thereby improve estimation

efficiency if kept unadjusted (Leone et al., 2014). Nevertheless, Newbold, Carlson and Thorne (2010)

present descriptive evidence of the problematic with extreme values, or outliers, and the winsorization

level is therefore set to 1%. Because of the debate concerning ad hoc data modifications, such as

winsorization (see Leone et al., 2014), the statistical analyses containing non-winsorized data is presented

in Appendix 3.

3.5.2 Correlation Analysis

Drawing on the research design and the quantitative characteristics of this thesis, statistical analyses of

variables are conducted. Firstly, bivariate analyses including all variables are executed in order to gain

additional information about the associations of two variables. As the majority of the variables are

continuous and parametric, i.e. ratios or intervals, the Pearson’s correlation coefficient is applied. The

correlation intends to measure the strength and direction of a linear relationship (Collis & Hussey, 2014).

Although, correlation analyses do not determine the causation of dependent and independent variables

respectively, it is of particular use for this thesis as it enables the authors to avoid issues related to

multicollinearity. Multicollinearity is the situation when two variables are highly correlated and could

thus damage the effect of multiple regressions (Blumberg, Cooper & Schindler, 2011).

18

In accordance with Collis and Hussey (2014), Pearson’s correlation analysis is complemented with

bivariate scatterplot analysis, which allows for additional interpretation of associations between two

variables visually (see Appendix 3). This enables the authors to study patterns of points in graphs, which

give an indication of the strength and direction of a linear relationship (Collis & Hussey, 2014) and

detects potential extreme values that could affect the correlation.

3.5.3 Multiple Regression Analysis

The Pearson’s correlation analysis is of importance when examining the correlation between two

variables. However the analysis fails to explain the correlation between several variables (Blumberg et al.,

2011). Instead, the multiple regression analysis is used as complement to examine the association between

the dependent variables and several independent variables. Thus, the multiple regression analysis allows

the user to explicitly control for several factors that simultaneously might affect the outcome of the

dependent variable (Wooldridge, 2013).

Multiple factors could affect the level of tone in CEO letters, and it is therefore advantageously to test

associations between variables through multiple regression analysis. H1 and H3 are tested using Ordinary

Least Square (OLS) regressions. The OLS regression enables the user to determine the effect of the

independent variables on the dependent variables whilst simultaneously controlling for factors that might

influence the association (Wooldridge, 2013). Hence, referring to H1 and H3, the OLS regression analysis

is considered an appropriate statistical analysis method based on the normally distributed dependent

variables. Furthermore, the model is considered particularly appropriate when identifying abnormal

values since the model generates an error term (ɛ), which count for the value unexplained by the control

variables and thus classifies as abnormal (Wooldridge, 2013).

The dependent variable in H2 is ABPOS, a dummy variable derived from ABTONE. The ABPOS variable

is further characterized as binary since the outcome of the variable is either 1 or 0, which disables the use

of OLS regression analysis. The distribution is further identified as binomial (Rabe-Hesketh & Everitt,

2004), and logistic regression analysis is therefore appropriate in order to test H2. Whilst the OLS

regression calculates the correlation between the dependent and independent variables, the logistic

regression calculates the probability of receiving a certain event of interest (Rabe-Hesketh & Everitt,

2004). The interpretation of the logistic regression model therefore differs from the OLS regression

model.

The issue with multicollinearity, which is often related to multiple regression analysis, is prevented

through conducting bivariate analysis (see section 3.5.2). Furthermore, in order to control for

heteroskedasticity, and increase the robustness of the findings, the robust standard error is included when

running the OLS regressions (Wooldridge, 2013). Due to the composition of the logistic regression

model, the heteroskedasticity issue is automatically counted for (Wooldridge, 2013).

19

3.5.3.1 Regression models Similar to previous tone studies (e.g. Davis & Tama-Sweet, 2012; Huang et al., 2014) the regressions of

this thesis are annual cross-sectional regressions of TONE, ABPOS and ABTONE on determinants

previously explained in section 3.3. Specifically, the regressions are6:

𝐻1: 𝑇𝑂𝑁𝐸 = 𝛽0 + 𝛽1𝐸𝐴𝑅𝑁 + 𝛽2𝑆𝐼𝑍𝐸 + 𝛽3𝐵𝑇𝑀 + 𝛽4𝑅𝐸𝑇 + 𝜀 (1)

𝐻2: 𝑃(𝐴𝐵𝑃𝑂𝑆 = 1) = 𝛽0 + 𝛽1𝐸𝐴𝑅𝑁 + 𝛽2𝐿𝑂𝑆𝑆 + 𝛽3𝑆𝐼𝑍𝐸 + 𝛽4𝐵𝑇𝑀 + 𝛽5𝑅𝐸𝑇 + 𝛽6𝐸𝐴𝑅𝑁_𝐿𝑂𝑆𝑆 + 𝜀 (2)

𝐻3: 𝐴𝐵𝑇𝑂𝑁𝐸 = 𝛽0 + 𝛽1𝐴𝐵𝐴𝐶𝐶 + 𝛽2𝐸𝐴𝑅𝑁 + 𝛽3𝐿𝑂𝑆𝑆 + 𝛽4𝑆𝐼𝑍𝐸 + 𝛽5𝐵𝑇𝑀 + 𝛽5𝑅𝐸𝑇 + 𝛽6𝐸𝐴𝑅𝑁_𝐿𝑂𝑆𝑆 + 𝜀 (3)

3.6 Data Collection

The collected data have both quantitative (numbers) and qualitative (narratives) characteristics (Collis &

Hussey, 2014). The quantitative data refers to the financial figures needed to measure earnings

management and tone management, and is collected from the database Datastream, provided by Thomson

Reuters. The variables used to collect firm information from Datastream are provided in Appendix 4. Due

to the limitation of Datastream, the sample firms are collected through MiFID Database, provided by the

European Securities and Markets Authority (ESMA). The version date is set to 31-12-2013 in order to

capture the firms listed on the London Stock Exchange by the end of year 2013.

The qualitative data refer to the measurement of tone, which is executed using the CEO letters of annual

reports (see section 3.4.1), as annual reports are accessible and directed to the firm’s most important

audience, i.e. its stakeholders (Hildebrandt & Snyder, 1981). Further on, the textual information will be

translated into numerical form in order to determine the tone variable. In other words, the qualitative data

will transform into quantitative.

3.7 Reliability and Validity

In general terms, quantitative studies possess high levels of reliability and relatively low levels of validity

(Collis & Hussey, 2014). The reliability of this thesis is strengthened by the providence of the word list

and sample used to calculate tone (see Appendix 1 and 2), which otherwise could affect the degree of

reliability.

To enhance the validity of this thesis, several aspects are taken into consideration throughout the

execution phase. Firstly, only CEO letters or equivalent letters signed by the CEO or the President are

included in the final sample in order to increase validity and secure consistent application. Secondly,

equal amount of positive and negative words is used to ensure that no category is overly represented.

Also, to the extent possible, the power of the words balances each other (for example fortunately and

unfortunately). Thirdly, regarding certain words, the spelling of both American and British English is

acknowledged with the purpose to capture these words. Finally, as stated in section 3.4.1, if negations

appear immediately before a positive word, the positive word is characterized as negative.

6 Industry variables are included in all regressions, but not reported.

20

3.8 Limitations

Additionally, it is acknowledged that this thesis involves limitations. Firstly, the classification of positive

and negative words partly involves judgments made by the authors. Moreover, the meaning of some

words might not be fully captured since the meaning is determined by the context. For example, the

directional words increase and decrease could appear in both positive and negative circumstances due to

their ambiguous nature. However, Henry (2008) provides results confirming that increase (decrease) in

majority appears in a positive (negative) context. Despite its limited nature, the approach is commonly

applied and thereby considered appropriate for the purpose of this thesis.

Finally, a final sample of 415 firms can be perceived as small in order to make generalizations about a

population. Furthermore, the sample size is considered large enough to address the research problem

(Collis & Hussey, 2014), and test the hypotheses.

21

4. Empirical Results and Analysis

This chapter presents the empirical results of the statistical tests performed in this thesis, and their

implications for the tested hypotheses. The introductory part of the chapter presents descriptive statistics,

which aim to present a numerical summary of the variables used in the statistical analysis. The

introductory part is followed by a presentation of the correlation analysis, which thereafter concludes in

the multiple regression analysis. Throughout the chapter, empirical results are analyzed with regards to

previous literature.

4.1 Descriptive Statistics

Table 4.1.1 Descriptive Statistics Number of observations: 415

Variable Mean Median Std.Dev P1 P99 Skewness Kurtosis

TONE 0.6560 0.6761 0.2056 -0.2121 1 -0.9717 4.5327

EARN* 0.0480 0.0473 0.0899 -0.3390 0.2686 -1.0622 7.7633

SIZE 5.9124 5.9307 0.9034 3.2646 8.2393 0.0104 3.0090

BTM* 0.6822 0.5176 0.6143 -0.1545 3.5975 2.1637 9.6696

RET** 0.2691 0.2411 0.4339 -0.8071 2.0455 0.9811 6.0660

LOSS 0.1590 0 0.3662 0 1 0.1341 4.4770

ABTONE -1.06e-10 0.0109 0.1730 -0.5131 0.4799 -0.6743 4.2268

ABACC 0.9353 0.4047 4.4958 0.0002 87.5967 17.5240 334.9104

ABPOS 0.5277 1 0.4998 0 1 -0.1110 1.0123

* Winsorized at the 1% level in order to minimize the impact of outliers

** Winsorized at the 1% level, high only, in order to minimize the impact of outliers

The table above presents descriptive statistics regarding the variables included in the statistical tests. The

TONE variable results in a mean (median) of 65.60% (67.61%), indicating that the total level of tone in

CEO letters is generally positive. The 99 percentile present a value of 1, which implies the occurrence of

no negative words among the sample firms. Huang et al. (2014) present a mean (median) of 43% (42%),

which corresponds to the figures of Henry (2008) who presents a mean (median) of 56.80% (60%). The

descriptive statistics in this thesis are mainly compared to the descriptives of Henry (2008) due to the

similarities in measuring tone. Hence, the descriptive statistics (table 4.1.1) reveal resemblance with

previous literature and confirms that the general level of tone in CEO letters is similar to the general tone

in earnings press releases. Furthermore, the ABTONE variable reveals a somewhat negative mean value

and a somewhat positive median value. However, low values are expected since ABTONE is a regression

residual.

As described in section 3.5.1, some variables are winsorized in order to fulfill the OLS regression

requirement of normal distribution. The raw data, i.e. the non-winsorized values are presented in

Appendix 3.

22

Table 4.1.2 Binary Variables

Variable Count

ABPOS

1 = ABTONE > 0 219

0 = ABTONE ≤ 0 196

TOTAL 415

LOSS

1 = EARN < 0 66

0 = EARN ≥ 0 349

TOTAL 415

The above table presents the distribution of the binary variables. The numbers reveal that just over half of

the sample firms employ abnormal positive tone according to the chosen proxy (ABTONE >0). The table

is not involved in the hypotheses tests, but contributes to the discussion in chapter five.

4.2 Correlation Analysis

Table 4.2 Pearson’s Correlation Analysis

Number of observations: 415

TONE EARN SIZE BTM RET LOSS ABTONE ABACC ABPOS

TONE 1

EARN 0.2969* 1

(0.0000)

SIZE 0.3787* 0.2313* 1

(0.0000) (0.0000)

BTM -0.3721* 0.4034* -0.4444* 1

(0.0000) (0.0000) (0.0000)

RET 0.2483* 0.3297* 0.0138 -0.3406* 1

(0.0000) (0.0000) (0.7795) (0.0000)

LOSS -0.2535* -0.6714* -0.2710* -0.3134* -0.2217* 1

(0.0000) (0.0000) (0.0000) (0.0000) (0.0000)

ABTONE 0.8414* 0.0000 0.0000 -0.0000 0.0000 -0.0291 1

(0.0000) (1.0000) (1.0000) (1.0000) (1.0000) (0.5540)

ABACC 0.0114 0.0448 -0.0193 0.0489 -0.0326 -0.0415 0.0400 1

(0.8165) (0.3630) (0.6945) (0.3206) (0.5084) (0.3991) (0.4158)

ABPOS 0.6242* -0.0395 -0.0542 0.0202 -0.0394 0.0155 0.7694* 0.0277 1

(0.000) (0.4224) (0.2708) (0.6818) (0.4229) (0.7536) (0.0000) (0.5734)

* Indicate significance at the 5% level

23

The purpose of the Pearson’s correlation analysis is as described in section 3.5.2 to explore the correlation

between two variables, and further to detect issues of multicollinearity before conducting the multiple

regression analysis. The correlation coefficient expresses the relationship between two variables given

different levels of significance (Blumberg, 2011). The intercorrelations are presented in table 4.2 and

reveal no indication of multicollinearity, as the majority of the correlation coefficients are considered low

(Collis & Hussey, 2014)7. Although, the variables TONE, ABTONE and ABPOS are arguably highly

intercorrelated, however this result is expected since ABTONE is the residual from regressing TONE, and

ABPOS is a dummy variable generated from ABTONE. The issue of multicollinearity is thus not present

since the variables are not used simultaneously in the regressions.

The correlation between LOSS and EARN (-0.6714) is according to Collis and Hussey (2014) interpreted

as medium and the question to include both variables simultaneously in the multiple regression analysis

therefore arises. This potential issue is considered by including the interaction variable EARN_LOSS in

regressions (2), (3) where both variables are used simultaneously. Furthermore, Collis and Hussey (2014)

state that variables with low or medium correlation coefficients will not create bias in subsequent

statistical analyses.

4.3 Multiple Regression Analyses

The below tables present the empirical data resulting from the multiple regression analyses. The aim of

the regression on total accruals (see 3.4.2) is to collect the variable ABACC, and the regression is

therefore not presented in this context. The dummy variable ICB is included in all regressions, but not

further reported in tables 4.3 (I), (II). The p-values are used in the decision to accept or reject the null

hypotheses regarding both the OLS regressions and the logistic regression. The alternative hypotheses are

considered supported if p < the level of significance. The regression models used in the statistical tests are

presented below.

𝐻1: 𝑇𝑂𝑁𝐸 = 𝛽0 + 𝛽1𝐸𝐴𝑅𝑁 + 𝛽2𝑆𝐼𝑍𝐸 + 𝛽3𝐵𝑇𝑀 + 𝛽4𝑅𝐸𝑇 + 𝜀 (1)

𝐻2: 𝑃(𝐴𝐵𝑃𝑂𝑆 = 1) = 𝛽0 + 𝛽1𝐸𝐴𝑅𝑁 + 𝛽2𝐿𝑂𝑆𝑆 + 𝛽3𝑆𝐼𝑍𝐸 + 𝛽4𝐵𝑇𝑀 + 𝛽5𝑅𝐸𝑇 + 𝛽6𝐸𝐴𝑅𝑁_𝐿𝑂𝑆𝑆 + 𝜀 (2) 𝐻3: 𝐴𝐵𝑇𝑂𝑁𝐸 = 𝛽0 + 𝛽1𝐴𝐵𝐴𝐶𝐶 + 𝛽2𝐸𝐴𝑅𝑁 + 𝛽3𝐿𝑂𝑆𝑆 + 𝛽4𝑆𝐼𝑍𝐸 + 𝛽5𝐵𝑇𝑀 + 𝛽5𝑅𝐸𝑇 + 𝛽6𝐸𝐴𝑅𝑁_𝐿𝑂𝑆𝑆 + 𝜀 (3)

7

(-)0.90 to (-)0.99 (very high correlation)

(-)0.70 to (-)0.89 (high correlation)

(-)0.40 to (-)0.69 (medium correlation)

0 to (-)0.39 (low correlation)

24

Table 4.3 Multiple Regression Analyses

VARIABLES TONE (1) ABPOS (2) ABTONE (3)

EARN 0.3075* -1.7564 -0.2084

(0.022) (0.364) (0.213)

SIZE 0.0638** -0.0858 -0.0037

(0.000) (0.515) (0.728)

BTM -0.0394 -0.0123 0.0078

(0.070) (0.955) (0.646)

RET 0.0579** 0.2613 0.0029

(0.007) (0.332) (0.888)

LOSS -0.1413 -0.0151

(0.718) (0.722)

EARN_LOSS 0.5808 0.3764

(0.858) (0.275)

ABACC 0.0015**

(0.0005)

Observations 415 415 415

R-squared 0.2921 0.0102 0.0082

* Indicate significance at the 5% level

** Indicate significance at the 1% level

Robust p-statistics in parentheses

Industry effects are included in all regressions but not reported

4.3.1 Hypothesis Testing

The purpose of applying multiple regression analysis is, for this thesis, to empirically test the previously

presented hypotheses. The results are interpreted and analyzed in the sections below and the data referred

to is presented in tables 4.3 (above). Furthermore, the support, or lack of support, of the hypotheses is

summarized in table 4.3.1.

The R2 value, i.e. the square of the multiple correlation coefficients, expresses the strength of the linear

relationship between the independent variables and control variables and the dependent variable

(Newbold et al., 2010). Hence, the R2 explains the fit between the variables and the model, and thus to

what extent the dependent variable is explained by the independent variables and control variables

(Newbold et al., 2010). The R2 value will be commented on, however not further discussed, in relevant

sections.

4.3.1.1 Tone and Financial Performance The correlation coefficients resulting from regression (1) reveal, as expected, a positive association

(0.3075) between TONE and EARN, and the p-value (0.022) indicates significant association at the 5%

level, after controlling for firm fundamentals. Thereby, the empirical results provide evidence to support

H1, and H0 is therefore rejected. Significant positive associations are also presented regarding SIZE and

RET, which indicates that the association between tone and firm fundamentals to great extent is positive.

The exception is BTM, which display a negative and insignificant association with TONE. The square of

25

the multiple correlation coefficient (R2) is 0.2921, indicating that 29.21% of TONE is explained by EARN,

SIZE, BTM and RET.

Interpretation of the regression results reveals that the tone in CEO letters can be derived from financial

performance, company size and annual stock return. Considering financial performance, the association

confirms the expected outcome and present empirical evidence of a significantly positive association

between TONE and EARN, indicating that the level of tone in CEO letters increases in line with firm

performance. The direction and magnitude is further strengthened by the Pearson’s correlation analysis,

which presents a positive and significant correlation (0.2969). Furthermore, in accordance with Davis and

Tama-Sweet (2012), the empirical results present firm performance to be the main determinant of tone

due to its relative high explanatory power. Moreover, scholars (e.g. Amernic & Craig, 2007; Merkl-

Davies & Brennan, 2007; Henry, 2008) argue that when profits are high, management tend to use a

positive language as a result of good management, which also support the findings of a positive

correlation between firm performance and tone.

More so, the positive and significant correlation between firm size and tone indicate that tone increases

with the size of the firm. Courtis (1998) argues that firms subject to media attention might have motives

to influence readers’ perceptions, which could explain the positive correlation. In contradiction to the

empirical findings, Li (2010) and Huang et al. (2014) present an inverse relationship between tone and

size, indicating that larger firms employ less positive tone. Based on previous research, Li (2010) argues

that larger firms might have incentives to be cautious in their expressions to avoid political costs, and thus

moderate the tone. Nevertheless, the Pearson’s correlation analysis (table 4.2) presents a positive and

significant correlation of 37.87%, which when controlling for firm fundamentals is still significant,

although lower.

The control variables representing future growth opportunities (RET and BTM) present different empirical

results. Similar to firm size, the RET coefficient reveal a significant positive association between stock

returns and tone in CEO letters. These results indicate that the tone increases in line with stock returns,

which is also confirmed by Li (2010). On the contrary, the BTM coefficient reveals a negative association.

However, due to the insignificance of the correlation, the effect of BTM on tone in CEO letters cannot be

determined. Nevertheless, when not controlling for firm fundamentals, the Pearson’s correlation analysis

reveals a negative and significant correlation between tone and BTM, which is consistent with the findings

of Huang et al. (2014). Implications for BTM are further discussed in chapter five.

4.3.1.2 Abnormal Positive Tone and Financial Performance H2, i.e. the probability of observing abnormal positive tone as financial performance decreases, is tested

through logistic regression analysis. The empirical results, table 4.3 and regression (2), present negative

β-values, which indicate that an increase in the independent variables result in decreased probability that

the event will occur. However, the results are not significant at conventional levels and H2 can thus not be

considered supported, leading to acceptance of H0. The R2 value is 0.0102, revealing that ABPOS is to

1.02% explained by the independent and control variables.

Huang et al. (2014) state that management abuses abnormal positive tone when firm fundamentals are

poor. Furthermore, Huang et al. (2014) explain that quantitative information can be used to signal better

future prospects, which also aligns with Hildebrandt and Snyder (1981) and Rutherford (2005).

26

Specifically, Hildebrandt and Snyder (1981) imply that years with poor performance include positive

statements, i.e. the tone is inconsistent with true financial performance resulting in abnormal positive

tone. Despite these indications of a relation between abnormal positive tone and financial performance,

the insignificant empirical results of this thesis cannot justify such documentation in the CEO letter.

4.3.1.3 Abnormal Tone and Abnormal Accruals Regression (3) aims to test the relationship between the abnormal tone in CEO letters and abnormal

accruals (H3). The alternative hypothesis states a positive association, meaning that tone management and

earnings management are expected to be present simultaneously. The empirical results present positive

correlation (0.0015) between ABTONE and ABACC, after controlling for firm fundamentals. The p-value

of the correlation (0.0005) reveals statistical significance of the correlation between the two variables.

However, table 4.3 reveals no other correlations in regression (3) to be statistically significant. Still, the

association of interest is significant and positive, and H0 is therefore rejected in favor of H3. The R2 value

of regression (3) is 0.0082, which is considered low but yet accepted.

Although the correlation coefficient (0.0015) could be considered low, Huang et al. (2014) explain that a

low correlation between abnormal accruals and abnormal tone might be because of estimation errors. This

explanation could be considered valid in this case since both abnormal accruals and abnormal tone are

residual values. Nevertheless, the correlation is highly significant (0.0005), and the results can thus be

used to reject the null hypothesis.

Thereby, as anticipated, the statistical findings present a positive significant association between

abnormal tone in CEO letters and abnormal accruals, which in turn supports the main purpose of this

thesis. The findings are in accordance with Huang et al. (2014) who provide similar result in earnings

press releases. Drawing on the results presented in table 4.3 along with the literature, the positive

association indicates that tone management and earnings management co-occur as complements, and are

thus not substitutes. Furthermore, Schrand and Walther (2000) touch upon the co-occurrence of tone

management and earnings management as they imply that strategic disclosure and earnings management

are interrelated. As earnings management could occur everywhere in financial reporting (Schipper, 1989),

the CEO letter evidently functions as a channel for strategic reporting based on the presented empirical

findings.

The presented association is further supported by the similar incentives behind the two management

behaviors, for instance, the desire to mislead stakeholders by misrepresent firm performance (Leuz et al.,

2003; Li, 2008; Huang et al., 2014). Li (2010) expects the association between tone and accruals to be

positive when incentives to mislead investors are present. Considering the purpose to test abnormality, the

positive correlation between abnormal tone and abnormal accruals could thus indicate the presence of

incentives to manipulate users’ perceptions.

27

Table 4.3.1 Summary Hypotheses tests

H1 Tone in CEO letters is positively associated with financial performance

Supported

H2 The probability of observing abnormal positive tone in CEO letters increases as financial

performance decreases Not supported

H3 Abnormal tone in CEO letters is positively associated with abnormal accruals Supported

28

5. Discussion

The chapter includes discussions surrounding the role of the CEO letter with respect to the empirical

findings and analyses presented in chapter four, together with personal reflections made by the authors.

Additionally, the chapter evolves from the introducing chapters of this thesis concerning the issues

surrounding the subject. Finally, the chapter serves as foundation to the concluding remarks in following

chapter.

Throughout the thesis, the CEO letter is argued to be an important document for users of annual reports.

Several scholars state that the letter is a significant narrative document through which management have

opportunities to express beliefs and values to their stakeholders. For instance, Huang et al. (2014) state

that accounting narratives are important in order to understand quantitative information. Furthermore,

scholars state the letter to be informative, since it addresses past and future events (McConnell et al.,

1986). However, whether the information is useful or not has been discussed in prior literature.

McConnell et al. (1986) and Abrahamson and Amir (1996) argue for its usefulness as the content of the

CEO letter could be used to assess future performance. However, one could argue that such letters are

only completely useful when they possess a neutral tone, i.e. when tone reflects firm fundamentals. Thus,

when management employ abnormal tone in CEO letters, information asymmetry might increase since

users are forced to read between the lines in order to hold accurate perceptions about the firm.

Consequently, when collection of information is obstructed, earnings forecasts, investment decisions and

other stakeholder actions might be biased and the usefulness of the CEO letter thus decreases.

The findings presented in table 4.3 regression (2), reveal insignificant results, yet negative coefficients.

Due to the insignificant results, no further conclusions can be drawn regarding the probability of

observing abnormal positive tone when financial performance decreases. The outcome is unfortunate

since the directions of the coefficients align with statements made by previous scholars. Nevertheless,

based on these statements (see Hildebrandt & Snyder, 1981; Rutherford, 2005; Huang et al., 2014; Tan et

al., 2014), one could believe that firms with decreased financial performance are more prone to engage in

tone management. In fact, table 4.1.2 reveals that 219 firms employ abnormal positive tone. This together

with the results in table 4.3 indicates that abnormal positive tone exists, but might be explained by other

factors than included in the regression model (2). Regardless of the causes behind abnormal positive tone,

one could advocate it to be problematic since firm fundamentals would be misrepresented. Consequently,

in such cases, the information provided in the CEO letter would conflict with basic financial premises,

such as true and fair view, and lower the quality of the reported content. Engagement in tone management

is therefore considered problematic. Moreover, Huang et al. (2014) document a reversal effect in future

earnings when abnormal positive tone is present. The authors conclude the reversal effect to result from

overestimated reactions to abnormally positive earnings announcements, which points towards the

drawbacks with employing abnormally positive tone. The drawbacks are also stressed in the research by

Tan et al. (2014), who state the risk of positive language to “backfire”.

Considering the negative subsequent market reaction, why do firms engage in tone management? One

answer could be short-term thinking. The reversal effect reveals positive market reactions immediately

after the earnings announcement, which could explain tone management behavior. More so, firms might

find it necessary to contemporary satisfy investors through employing abnormal positive tone and signal

29

good future prospects. As with the free nature of the CEO letter, along with studies revealing a positive

reaction to distorted statements about true financial performance, incentives to manage tone in CEO

letters when profits are down, are arguably present. Another answer to the use of tone management could

be attribute framing, i.e. when users’ perceptions about identical items differ depending on the tone used

to describe certain attributes. Hence, if management possesses this knowledge, attribute framing creates

an incentive to use optimism in CEO letters as the reader would perceive performance to be better than

indicated by financial figures.

Tone could also be negatively managed, i.e. abnormal negative tone. To clarify, this is the case when firm

performance is positive, but the tone is downwardly managed. Incentives to decrease the level of tone in

CEO letters could be present when firms are subject to political litigation costs or being accused of

monopolistic behavior, which mainly involve large firms. The empirical findings reveal a negative

association between firm size and abnormal tone. However, these findings are not statistically significant,

and one could therefore not draw any conclusions based on the results since a potential alternative

hypothesis would be unsupported. Additional reason to downwardly bias the tone could be the growth

opportunities of the firm. Considering Li (2010), it could be assumed that fast growing firms, facing

uncertain economic environment are cautious in their disclosures, and thus less positive. On the other

hand, with respect to Huang et al. (2014), who propose the opposite, one may also assume that a fast

growing firm is optimistic about the future, causing it to employ a more positive tone.

Aligned with incentives to employ abnormal tone, incentives behind earnings management behavior

reasonably resemble. With inspiration from Leuz et al. (2003) and Dechow et al. (2010), one may

interpret that there are incentives to engage in earnings management in order to distort users’ perceptions

when firm performance is poor. Based on the literature along with the empirical findings, firms may

through the CEO letter provide positive statements, commenting on their excellent sales for instance,

although these could be mainly driven by an abnormal amount of accruals. In this case, neither the tone

nor the revenues reflect true firm fundamentals. Thus, this could be considered the case where the two

management behaviors complement each other. The positive association between abnormal tone and

abnormal accruals is justified through empirical findings, and this reasoning about the two behaviors is

thus valid.

Referring to the discussion about the usefulness of the CEO letter, the co-occurrence of tone management

and earnings management reasonably lowers the usefulness of both the CEO letter and reported earnings.

Abnormal accruals have been defined to possess low earnings persistence, and thus low earnings quality,

which further results in decreased information usefulness. Furthermore, Li (2010) states that a positive

association between tone and accruals indicates that management has incentives to report strategically.

Hence, when including abnormality, one could expect the co-occurrence of the two behaviors to be

opportunistic, and that management is aware of the decreased information quality. More so, the intention

to change users’ perceptions about firm fundamentals could be considered present when tone

management and earnings management operate jointly.

However, both the occurrence of tone management and earnings management could, according to

previous research, be unintentional. For example, Huang et al. (2014) mention that tone management

could be applied for informative purposes, i.e. when firm fundamentals are better than presented by

30

financial figures. One could also assume that due to complexity or uncertainties in operations,

management apply higher or lower tone in order to explain certain events, excluded from the financial

reporting, and thus reduce information asymmetry. Drawing on the same reasoning, complexity in

business operations or uncertainties in the business environment could increase the level of accruals due

to the requirements of estimation of future events, and the composition of the accounting regulation.

Thereof, the positive association between tone management and earnings management could be of

informative purposes rather than strategically purposes. Nevertheless, throughout the thesis and in line

with previous literature (see Courtis, 1998; Li, 2010; Huang et al., 2014), the intentions behind

abnormality in tone and accruals are considered strategic. Hence, the jointly occurrence of tone

management and earnings management is considered to depend on strategic purposes rather than

informative purposes.

31

6. Concluding Remarks

This chapter aims to present concluding remarks with regards to previously presented chapters, the

purpose and stated hypotheses.

The main purpose throughout this master thesis is to establish whether there is a positive association

between tone management in CEO letters and earnings management in firms listed on the London Stock

Exchange during year 2013. The additional purpose is to establish the potential association between tone

in CEO letters and financial performance. In the execution phase, the authors develop a customized

dictionary to fit with the stated purpose, which subsequently enables the authors to determine the level of

tone in CEO letters. The execution phase continues in correlation analysis and multiple regression

analysis in order to determine the support, or lack of support, of the hypotheses. Based on the results

accomplished in this study, several conclusions can be established along with strengthened contribution to

accounting theory within the area of accounting narratives, a field that is still emerging and thus relatively

underexplored.

Overall, the empirical analyses in this thesis present mixed findings with regards to initial expectations.

Referring to the descriptive statistics, the tone in CEO letters appears to be generally positive according to

the mean and median values. Given the role of the CEO letter and the stated fact that it is unregulated, it

is not surprising that the tone appeared generally positive among the sample firms. Furthermore, the CEO

letter is, according to previous scholars, the most read section of the annual report and previous research

establish that the tone has an impact on investors and other users of the annual report. It is therefore

credible that management uses this section to present positive statements about the firm. Thus, the

importance of the CEO letter justifies both the generally positive tone and the empirical findings of a

positive association between tone and financial performance. In conclusion, the CEO letter is established

as a medium through which management communicate increased financial performance using positive

tone. Furthermore, the empirical results reveal that the tone in CEO letters is derived, besides financial

performance, from company size and annual stock return.

Finally, the findings reveal that abnormal tone and abnormal accruals are positively associated. This

empirical finding argues for a situation where both tone management and earnings management are used

simultaneously towards an opportunistic goal, to misrepresent firm fundamentals by managing tone in

CEO letters and altering financial figures. Based on the empirical findings and the excessive room for

opportunism, the general conclusion is that firms employ tone management in CEO letters to hide

earnings management behavior. Hence, accounting narratives may not always better indicate firm

fundamentals. This conclusion strengthens the fact that investors and other users of the annual reports

should take a “hard look” at the qualitative information along with the quantitative information and

consider the content before creating a general perception of the firm. Consequently, the findings further

strengthen the role of the tone in CEO letters and highlight its influence when evaluating financial

performance.

32

7. Suggestions for Further Research

This chapter aims to present suggestions for other research opportunities in accounting research within

accounting narratives. As the area of lexical features is relatively underdeveloped, the suggestions mainly

regard this area with respect to different approaches to capture and classify linguistic features in

corporate communication channels.

The area of lexical features within accounting research is evolving, yet still developing. Due to the

relatively late focus on accounting narratives, there are several opportunities for further research. This

thesis targets tone management in terms of positive versus negative words in the CEO letter. Future

studies could therefore concern other lexical features such as certain or uncertain language together with

abnormal accruals. For example, abnormal accruals might be reported due to uncertain business

environment. A language classified as “certain” would thus not be consistent with firm fundamentals and

indicate abnormal use of language. Furthermore, extending the contributions of this thesis, one suggestion

is to include additional control variables such as business complexity and firm age.

In addition, this thesis includes firms listed on the London Stock Exchange, suggestion for further

research is therefore to incorporate other sample criteria and potentially compare the usage of tone

management between different markets. Specifically, a comparison between UK listed firms and US

listed firms could be interesting due to differences in accounting regulations. Also, inspired by Leuz et al.

(2003), a further suggestion is to cluster firms in different groups, e.g. country, and study differences in

language use between these groups.

Studying if annual reports are readable can further capture linguistics in corporate reports. Annual report

readability refers to reading ease and concerns the extent to which accounting information is textually

and/or visually readable. For example, Li (2008) finds that firms with less persistent earnings present

complex disclosures with low readability. Tan et al. (2014) on the other hand study the co-occurrence of

annual report readability and tone management in earnings press releases. As this thesis provides

evidence that strengthens the important role of the CEO letter, a suggestion is to establish if abnormal

tone and low readability are present simultaneously in CEO letters.

This master thesis is delimited to accounting narratives in CEO letters. On the other hand, the concept of

impressions management is explained by Clatworthy and Jones (2006) as the tendency of individuals, or

organizations, to selectively present data through graphics of narratives in order to positively influence

the recipient's’ perceptions about the presented information. With respect to impressions management, a

suggestion for further research is to consider the information presented in various management

communication channels together with the graphs surrounding the information. More specifically, a

suggestion for further research is to investigate the use of impressions management in CEO letters

together with the use of tone management.

33

List of References

Books:

Amernic, J., Craig, R., Tourish, D. (2010). Measuring and Assessing Tone at the Top Using Annual

Report CEO letters. Edinburgh: Institute of Chartered Accountants of Scotland. ISBN: 987-1-

904574-55-2

Anderson, D. R., Sweeney, D.J., Williams, T.A., Freeman, J., Shoesmith, E. (2009). Statistics for

Business and Economics, 10th edition, South-Western, Cengage Learning. ISBN: 978-1-84480-313-2

Blumberg, B. F., Cooper, D.R., Schindler, P.S. (2014) Business Research Methods, 4th edition,

McGraw-Hill Education Europe. ISBN: 978-0-077-15748-7

Collis, J., Hussey, R. (2014). Business Research: A Practical Guide for Undergraduate and

Postgraduate students, 4th edition, Palgrave Macmillan. ISBN: 978-0-230-30183-2

Newbold, P., Carlson, W. L., Thorne, B. (2010). Statistics for Business and Economics, 7th edition,

Pearson Education Limited. ISBN: 978-0-13-507248-6

Rabe-Hesketh, S., Everitt, B. S. ( 2004). A Handbook of Statistical Analyses using Stata, 3rd edition,

CRC Press LLC. ISBN: 1-58488-404-5

Wooldridge, J. M. (2013). Introductory Econometrics: A Modern Approach, 5th edition, South-

Western, Cengage Learning. ISBN: 978-1-408-09375-7

Academic Articles:

Abrahamson, E., Amir, E. (1996). The Information Content of the President’s Letter to Shareholders.

Journal of Business Finance and Accounting, Vol. 23, No. 8, 1157-1182

Aerts, W. (2005). Picking Up the Pieces: Impression Management in the Retrospective Attributional

Framing of Accounting Outcomes. Accounting, Organizations and Society 30, 493-517

Amernic, J., Craig, R. (2007). Guidelines for CEO-speak: Editing the Language of Corporate

Leadership. Strategy and Leadership, Vol. 35, No. 3, 25-31

Burgstahler, D. C., Hail, L., Leuz, C. (2006). The Importance of Reporting Incentives: Earnings

Management in European Private and Public Firms. The Accounting Review, Vol. 81, No. 5, 983-

1016

Courtis, J. K (1998). Annual Report Readability Variability: Tests of the Obfuscation Hypothesis.

Accounting, Auditing and Accountability Journal 11 (4): 459-472

34

Clatworthy, M., Jones, J. J (2003). Financial Reporting of Good News and Bad News: Evidence from

Accounting Narratives. Accounting and Business Research 33 (3): 171-185

Clatworthy, M., Jones, J. J (2006). Differential Reporting Patterns of Textual Characteristics and

Company Performance in the Chairman’s Statement. Accounting, Auditing and Accountability

Journal 19 (4): 493-511

Craig, R. J., Brennan, N. M., (2012). An Exploration of the Relationship between Language Choice

in CEO Letters to Shareholders and Corporate Reputation. Accounting Forum, 36, 166-177

Davis, A. K., Tama-Sweet, I. (2012). Managers’ Use of Language Across Alternative Disclosure

Outlets: Earnings Press Releases versus MD&A. Contemporary Accounting Research, Vol. 29, No. 3,

804-837

Dechow, P. M., Skinner, D. J. (2000). Earnings Management: Reconciling the Views of Accounting

Academics, Practitioners, and Regulators. Accounting Horizons, Vol. 14, No. 2, 235-250

Dechow, P., Ge, W., Schrand, C. (2010). Understanding Earnings Quality: A Review of the Proxies,

Their Determinants and Their Consequences. Journal of Accounting and Economics 50, 344-401

Dechow, P., Sloan, R. G., Sweeney, A. P (1995). Detecting Earnings Management. The Accounting

Review, Vol. 70, No. 2, 193-225

Feldman, R., Govindaraj, S., Livnat, J., Segal, B. (2010). Management’s Tone Change, Post Earnings

Announcement Drift and Accruals. Review of Accounting Studies 15, 915-953

Fisher, S. A., Hu, M. Y. (1988). Does the CEO's Letter to the Shareholders Have Predictive Value?

Business Forum Vol. 14, No. 1, 22-24

Geiger, M. A., North, D. S. (2011). Do CEOs and Principal Financial Officers Take a “Bath”

Separately or Together?: An Investigation of Discretionary Accruals Surrounding Appointments of

New CEOs and PFOs. Academy of Accounting and Financial Studies Journal Vol. 15, No. 1, 1-30

Godfrey, J., Mather, P., Ramsay, A. (2003). Earnings and Impression Management in Financial

Reports: The Case of CEO Changes. ABACUS, Vol. 39, No. 1, 95-123

Hales, J., Kuang, X., Venkataraman, S. (2010). Who Believes the Hype? An Experimental

Examination of How Language Affects Investor Judgments. Journal of Accounting Research Vol. 49

No. 1, 233-255

Healy, P., Wahlen, W. (1999). A Review of Earnings Management Literature and Its Implications for

Standard Setting. Accounting Horizons, Vol. 13, No. 4, 365-383

35

Henry, E. (2008). Are Investors Influenced by How Earnings Press Releases are Written? Journal of

Business Communication, Vol. 45, No. 4, 363-407

Hildebrandt, H. W., Snyder, R. D. (1981) The Pollyanna Hypothesis in Business Writing: Initial

Results, Suggestions for Research. The Journal of Business Communication, Vol. 18, No. 1, 5-15

Huang, X., Teoh, S., Zhang, Y. (2014). Tone Management. The Accounting Review, Vol. 89, No. 3,

1083-1113

Jones, M. J. (1996). Readability in Annual Reports: Western Versus Asian Evidence - a Comment to

Textualize. Accounting, Auditing and Accountability Journal, Vol. 9, No. 2, 86-91

Kothari, S. P., Leone, A. J., Wasley, C.E. (2005). Performance Matched Discretionary Accrual

Measures. Journal of Accounting and Economics, Vol. 39, 163-197

Leuz, C., Nanda, D., Wysocki, P. (2003). Earnings Management and Investor Protection: An

International Comparison. Journal of Financial Economics Vol. 69, 505-527

Leone, A. J., Minutti-Meza, M., Wasley, C. E. (2014). Influential Observations and Inference in

Accounting Research. Simon Business School Working Paper No. FR 14-06.

Li, F. (2008). Annual Report Readability, Current Earnings and Earnings Persistence. Journal of

Accounting and Economics 45, 221-247

Li, F. (2010). The Information Content of Forward-Looking Statements in Corporate Filings - A

Naïve Bayesian Machine Learning Approach. Journal of Accounting Research, Vol. 48, No. 5, 1049-

1102

Loughran, T., McDonald, B. (2011). When is a Liability not a Liability? Textual Analysis,

Dictionaries, and 10-Ks. The Journal of FInance, Vol. 66, No. 1, 35-65

Merkl-Davies, D. M., Brennan, N. M. (2007). Discretionary Disclosure Strategies in Corporate

Narratives: Incremental Information or Impressions Management? Journal of Accounting Literature,

Vol. 27, 116-196

McConnell, D., Haslem, J. A. & Gibson, V. R. (1986). The President's Letter to Stockholders: A New

look. Financial Analysts Journal Vol. 42, No. 5, 66-70

Patelli, L., Pedrini, M. (2014). Is the Optimism in CEO’s Letters to Shareholders Sincere?

Impression Management Versus Communicative Action Curing the Economic Crisis. Journal of

Business Ethics, Vol. 124, 19-34

Richardson, S. (2003). Earnings Quality and Short Sellers. Accounting Horizons, 49-61

36

Rutherford, B. A. (2005). Genre Analysis of Corporate Annual Report Narratives: A Corpus

Linguistic-Based Approach. Journal of Business Communication, Vol. 42, No. 4, 349-378

Ruxton, G. D. (2006). The Unequal Variance t-test is an Underused Alternative to Student’s t-test

and the Mann-Whitney U test. Behavioral Ecology, Vol. 17, No. 4, 688-690

Schipper, K. (1989). Commentary on Earnings Management. Accounting Horizon, Vol. 4, No. 3, 91-

103

Schrand, C. M., Walther, B. R. (2000). Strategic Benchmarks in Earnings Announcements: The

Selective Disclosure of Prior-Period Earnings Components. The Accounting Review, Vol. 75, No. 2,

151-177

Sloan, R. G. (1996). Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about

Future Earnings? The Accounting Review, Vol. 71, No. 3, 289-315

Stubben, S. R. (2010). Discretionary Revenues as Measure of Earnings Management. The Accounting

Review, Vol. 85, No. 2, 695-717

Tan, H., Wang, E., Zhou, B. (2014). When the Use of Positive Language Backfires: The Joint Effect

on Tone, Readability, and Investor Sophistication on Earnings Judgments. Journal of Accounting

Research, Vol. 52, No. 1, 273-302

Teoh, S., Welch, I., Wong, T. (1998). Earnings Management and the Long-Run Underperformance

of Seasoned Equity Offerings. Journal of Financial Economics 50, 63-100

Thomas, J. (1997). Discourse in the Marketplace: The Making of Meaning in Annual Reports.

Journal of Business Communication, Vol. 34, No. 1, 47-66

Xie, H. (2001). The Mispricing of Abnormal Accruals. The Accounting Review 76, 357-373

Zimmerman, D. W., Zumbo, B. D. (1993). Rank Transformations and the Power of Student’s t Test

and Welch t’ Test for Non-Normal Populations with Unequal Variances. Canadian Journal of

Experimental Psychology, Vol. 43, No. 3, 523-539

Electronic Resources

FRC (2015) Guidance on the Strategic Report. https://www.frc.org.uk/Our-

Work/Publications/Accounting-and-Reporting-Policy/Guidance-on-the-Strategic-Report.pdf

(Accessed May 11, 2015)

Other References:

Runesson, Emmeli (2014). Capital Market Research in Financial Accounting. School of Business,

Economics and Law, Gothenburg University, Lecture Notes, 2014-02-16.

37

Appendix 1 Sample firms

3I GROUP PLC BG GROUP PLC

888 HOLDINGS PLC BHP BILLITON PLC

A.G. BARR PLC BIOQUELL

ABBOTT LABORATORIES BISICHI MINING PLC

ABERFORTH GEARED INCOME TRUST BLOOMSBURY

ACACIA MINING PLC BODYCOTE

ACAL PLC BOEING CO

AFREN PLC BOOKER GROUP PLC

AGA RANGEMASTER GROUP PLC BOVIS HOMES GROUP

AIR CHINA LIMITED BP PLC

ALTUS RESOURCE BRAEMAR SHIPPING

ALUMASC GROUP PLC BRITISH LAND COMPANY

AMEC FOSTER WHEELER BRITISH POLYTHENE

AMINEX PLC BRITVIC PLC

ANA HOLDINGS INC BT GROUP PLC

ANGLO AMERICAN PLC BTG PLC

ANGLO PACIFIC GROUP BUNZL PLC

ANGLOGOLD ASHANTI BURBERRY GROUP

ANITE PLC BWIN.PARTY DIGI

ANTOFAGASTA PLC CABLE & WIRELESS

APR ENERGY PLC CADOGAN PETROLEUM

AQUARIUS PLATINUM CAFFYNS PLC

ARM HOLDINGS PLC CAIRN ENERGY PLC

ASEANA PROPERTIES CALEDONIA INVESTMENT

ASHMORE GROUP PLC CANACCORD GENUITY

ASIA RESOURCE CAPE PLC

ASSOCIATED BRITISH CAPITAL & COUNTIES PROPERTIES PLC

ASTRAZENECA PLC CAPITAL & REGIONAL

AVEVA GROUP PLC CAPITAL DRILLING

AVOCET CAPITAL GEARING

AVON RUBBER PLC CARCLO PLC

BABCOCK INT'L GROUP CARILLION PLC

BAE SYSTEMS CARNIVAL PLC

BALFOUR BEATTY PLC CARPETRIGHT PLC

BARLOWORLD LTD CASH CONVERTERS

BARRATT DEVELOPMENTS CATERPILLAR INC

BEALE CENTAMIN PLC

BELLWAY PLC CENTAUR MEDIA PLC

BERENDSEN PLC CENTRICA PLC

BERKELEY GROUP CESC LIMITED

38

CHARLES STANLEY EASYJET PLC

CHEMRING GROUP PLC ELECTROCOMPONENTS

CHIME COMMUNICATIONS ELECTRONIC DP PLC

CITY OF LONDON ELEMENTIS PLC

CITY OF LONDON GR ENDEAVOUR INTL CORP

CLARKSON PLC ENQUEST PLC

COATS GROUP PLC ENTERPRISE INNS PLC

COBHAM PLC ENTERTAINMENT ONE

COLT GROUP ESSENTRA PLC

CONNECT GROUP PLC EUROPEAN ASSETS

CONSORT MEDICAL PLC EVRAZ PLC

COSTAIN GROUP PLC EXPERIAN PLC

CRANSWICK PLC F&C ASSET MGMT

CREIGHTONS PLC FENNER PLC

CRESTON PLC FIDESSA GROUP

CRODA INTERNATIONAL FILTRONIC PLC

CSR PLC FINDEL PLC

DAILY MAIL & GENERAL FIRST QUANTUM

DAIRY CREST GROUP FIRSTGROUP PLC

DAIRY FARM INT'L FLYBE GROUP PLC

DARTY PLC FORESIGHT 4 VCT PLC

DATANG INTL FRENCH CONNECTION GR

DCC PLC FULLER'S SMITH & TURNER PLC

DE LA RUE PLC G4S PLC

DEBENHAMS PLC GALLIFORD TRY PLC

DECHRA PHARMA GAMES WORKSHOP GROUP

DEE VALLEY GROUP PLC GEM DIAMONDS

DEV'T SECURITIES PLC GENEL ENE

DEVRO PLC GENERAL ELECTRIC CO.

DIAGEO PLC GKN PLC

DIALIGHT PLC GLAXOSMITHKLINE

DOMINO PRINTING GLENCORE PLC

DOMINO'S PIZZA GR GRAFTON GROUP PLC

DP WORLD LTD GRAINGER PLC

DRAGON OIL PLC GREENCORE GROUP PLC

DRAX GROUP PLC GREENE KING PLC

DRS DATA & RESEARCH H.R. OWEN PLC

DS SMITH PLC HALFORDS GROUP PLC

DUNELM GROUP PLC HALMA PLC

E2V TECHNOLOGIES HAMMERSON PLC

39

HARGREAVES LANSDOWN PLC JOHN MENZIES PLC

HARVEY NASH GRP PLC JOHNSON MATTHEY PLC

HAYNES PUBLISHING JPMORGAN ELECT PLC

HAYS PLC JPMORGAN EUROPEAN INVESTMENT TRUST PLC

HELICAL BAR PLC JUPITER FUND MANAGEMENT

HENDERSON EUROTRUST KAZ MINERALS PLC

HENDERSON GROUP PLC KELLER GROUP PLC

HERITAGE OIL PLC KENTZ CORP

HIKMA PHARMACEUTICAL KEYSTONE INVESTMENT

HILLSHIRE BRANDS CO KINGFISHER PLC

HILTON FOOD GROUP KOFAX LTD

HOCHSCHILD MINING KONAMI CORP

HOGG ROBINSON LADBROKES PLC

HOMESERVE PLC LAMPRELL PLC

HONDA MOTOR CO., LTD LAND SECURITIES

HONEYWELL INTERNATNL LATCHWAYS PLC

HONGKONG LAND HLDGS LAVENDON GROUP PLC

HOWDEN JOINERY LILLY (ELI) AND CO.

HUNTSWORTH PLC LIONTRUST ASSET MGT

ICAP PLC LONDON & ASSOCIATED

IG GROUP HLDGS LONDON STOCK EXCH

IMAGINATION TECH GRP LONDONMETRIC PROPERTY PLC

IMPERIAL TOBACCO GRP LONMIN PLC

INCHCAPE PLC LOOKERS PLC

INMARSAT PLC LOW & BONAR PLC

INT'L BUSINESS MACHS MACFARLANE GROUP PLC

INTERMEDIATE CAPITAL MAJEDIE INVESTMENTS

INTERN'L BIOTECHNOLO MALLETT PLC

INTERNATIONAL PERSONAL FINANCE PLC MAN GROUP PLC

INTERSERVE PLC MANAGEMENT CON

INTERTEK GROUP MANCH&LONDON INV TR

INTNL FERRO METALS MANDARIN ORIENTAL

INTU PROPERTIE MARKS & SPENCER

INVESTMENT COMPANY MARSHALLS PLC

ITE GROUP PLC MAVEN INCOME

ITV PLC MCBRIDE PLC

J SAINSBURY PLC MEARS

JAMES FISHER & SONS MEDUSA MINING LTD

JERSEY ELECTRICITY MEGGITT PLC )

JKX OIL & GAS PLC MICHAEL PAGE

40

MILLENNIUM PLAYTECH PLC

MITCHELLS & BUTLERS PLAZA CENTERS N.V.

MITIE GROUP PLC POLYMETAL INTER

MITSUBISHI CORP POLYUS GOLD INTER

MITSUBISHI ELECTRIC PORVAIR PLC

MOLINS PLC PPHE HOTEL

MONKS INVESTMENT PREMIER FARNELL PLC

MORGAN ADVANCED PREMIER FOODS PLC

MORGAN SINDALL PREMIER OIL PLC

MOSS BROS GROUP PLC PRIVATE EQUITY

MOTHERCARE PLC PROMETHEAN WORLD

MS INTERNATIONAL PLC PROVIDENT FINANCIAL GROUP

MURRAY INCOME TRUST PV CRYSTALOX SOLAR PLC

N BROWN GROUP PLC QINETIQ GROUP

NARBOROUGH PLANT QUARTO GROUP INC

NATIONAL EXPRESS GRP QUINTAIN ESTATES

NCC GROUP PLC RANDGOLD RESOURCES

NEXT PLC RATHBONE BROTHERS

NMC HEALTHCARE LLC RAVEN RUSSIA LTD

NORCROS PLC RECKITT BENCKISER GROUP PLC

NORTH ATLANTIC SMALL RECORD PLC

NORTHAMBER PLC REDROW PLC

NORTHGATE PLC REED ELSEVIER PLC

NTT DOCOMO INC. REGUS PLC

OCADO GROUP PLC RENISHAW PLC

OCEAN WILSONS RENOLD PLC

OIL COMP. LUKOIL JSC RENTOKIL INITIAL PLC

OPTOS PLC RICARDO PLC

OXFORD BIOMEDICA RIGHTMOVE PLC

OXFORD INSTRUMENTS RIO TINTO PLC

PACE PLC ROBERT WALTERS PLC

PARAGON GROUP OF COMPANIES

PLC ROLLS-ROYCE

PARITY GROUP PLC ROTORK PLC

PEARSON PLC ROYAL DUTCH SHELL

PENNON GROUP PLC RPC GROUP PLC

PERFORM GROUP LTD RUSPETRO PLC

PERSONAL ASSETS TR S & U PLC

PETRA DIAMONDS LTD SABMILLER PLC

PETROFAC LIMITED SAFESTORE HOLDINGS PLC

PFIZER INC SALAMANDER ENERGY PLC

41

SAVILLS PLC TATE & LYLE PLC WALKER CRIPS GROUP

SCHRODERS PLC TAYLOR WIMPEY PLC WEIR GROUP PLC

SDL PLC TELECITY GROUP PLC WH SMITH PLC

SECURITIES TRUST TELECOM PLUS PLC WHITBREAD PLC

SEGRO PLC TEMPLETON EMERGING WILLIAM HILL PLC

SENECA GLOBAL TESCO PLC WINCANTON PLC

SENIOR PLC TEX HOLDINGS PLC WM. MORRISON SUPERMT

SEPURA PLC THE RANK GROUP PLC WOLSELEY PLC

SERCO GROUP PLC THE SAGE GROUP PLC WORKSPACE GROUP PLC

SEVERN TRENT PLC THOMAS COOK GROUP PLC WS ATKINS PLC

SHANKS GROUP PLC THORNTONS PLC XAAR PLC

SHIRE PLC TONGAAT HULETT LTD XCHANGING PLC

SIG PLC TOPPS TILES PLC XP POWER LTD

SIGNET JEWELERS LTD TORAY INDUSTRIES ZHEJIANG EXPRESSWAY

SKY PLC TOSHIBA CORPORATION ZOTEFOAMS PLC

SMITH & NEPHEW PLC TOWN CENTRE SECS

SMITHS INDUSTRIES PLC TOYOTA MOTOR CORP

SOCO INT'L PLC TRAVIS PERKINS PLC

SONY CORP TRIBAL GROUP PLC

SOURCE BIOSCIENCE TRIFAST PLC

SPECTRIS PLC TRINITY MIRROR PLC

SPEEDY HIRE PLC TT ELECTRONICS PLC

SPIRAX-SARCO ENGINEERING PLC TUI TRAVEL PLC

SPIRENT COMM TULLOW OIL PLC

SPIRIT PUB CO TYMAN PLC

SPORTECH PLC UBM PLC

SPORTS DIRECT INTERNATIONAL

PLC UDG HEALTHCARE

ST IVES PLC UK MAIL GROUP

ST. MODWEN PROPS. ULTRA ELECTRONICS

STAGECOACH GROUP PLC UNILEVER PLC

STANDARD LIFE UNISYS CORPORATION

STHREE PLC UNITE GROUP PLC

STOBART GROUP LTD UNITED UTILITIES PLC

STV GROUP PLC VECTURA GROUP PLC

SVG CAPITAL PLC VERIZON COMMUNICATNS

SYNERGY HEALTH PLC VESUVIUS PLC

SYNTHOMER PLC VICTREX PLC

T CLARKE PLC VITEC GROUP PLC

TALKTALK TELECOM GROUP PLC VOLEX PLC

TARSUS GROUP PLC VP PLC

42

Appendix 2 Positive Words *Obtained from Henry (2008) word list

Above* Commit Exceeds* Highest*

Accelerate Committed Excellent* Highly

Accelerating Compelling Exciting Impress

Accomplish* Confidence Exclusive Impressed

Accomplished* Confident Expand Impressive

Accomplishes* Contribute Expand* Improve*

Accomplishing* Contributed Expanded Improved*

Accomplishment* Contribution Expanded* Improvement*

Accomplishments* Deep Expanding* Improvements*

Achieve* Definite* Expands* Improves*

Achieved* Delighted Expansion* Improving*

Achievement* Deliver* Extend Increase*

Achievements* Delivered* Extended Increased*

Achieves* Delivering* Extremely Increases*

Achieving* Delivers* Far Increasing*

Add Desire Favor Incredible

Added Desirable Favorably Invaluable

Advanced Effectively Favour Invigorate

Ahead Efficient Favourably Invigorated

Appeal Efficiently Firm Larger*

Appealing Elevate Firmer Largest*

Attracted Elevated First Leader*

Attractive Encouraged* Flexibility Leading*

Attractively Encouraging* Fortunately Maximise

Augmenting Encouragingly Front Maximising

Award Engage Gain Maximize

Awarded Enhance Good* Maximizing

Beat* Enhanced Greater* Maximum

Beats* Enhancement Greatest* More*

Beating* Enjoy* Greatly Most*

Beneficial Enjoyed* Grew* Opportunities*

Benefit Enjoying* Grow* Opportunity*

Benefited Enjoys* Growing* Optimal

Best* Enthusiasm Grown* Optimally

Better* Enthusiastic Grows* Optimise

Boost Essential Growth* Optimising

Boosted Ever Healthy Optimism

Broaden Exceed* Heightened Optimistic

Certain* Exceeded* High* Optimize

Certainty* Exceeding* Higher* Optimizing

43

Optimum Satisfying Very

Outperform Sharp Wealth

Outperformance Sharpening Welcome

Outperformed Sharply Welcoming

Outstanding Solid* Well

Paramount Stimulate Well-placed

Passionate Stimulated Well-run

Pleased* Stimulates Win

Pleasing Stimulus Won

Positive* Strength*

Positively Strengths*

Positives* Strengthen*

Prestigious Strengthened*

Pride Strengthening*

Privilege Strengthens*

Profitable Strong*

Progress* Stronger*

Progressing* Strongest*

Progression Substantial

Promising Substantially*

Proud Succeeded*

Proudly Succeed*

Quickly Succeeding*

Raise Succeeds*

Raised Successful*

Rapid Successfully

Rapidly Success*

Record* Successes*

44

Negative words

Apprehensive Depressed* Eroding Issues

Ascend Depressive Erosion Lack

Ascendance Deteriorate* Error Lackluster

Ascendancy Deteriorated* Fail* Laggard

Austerity Deteriorates* Failing* Least*

Bad Deteriorating* Fails* Less*

Badly Difficult* Failure* Limit

Barriers Difficulties* Fall* Limited

Below* Difficulty* Fallen* Limits

Bias Dilutive Falling* Lose

Biased Diminish Falls* Loss

Biases Diminished False Lost

Cautious Diminishes Fear Low*

Challenge* Diminishing Fearing Lower*

Challenged* Disappoint* Fears Lowered*

Challenges* Disappointed* Fell* Lowest*

Challenging* Disappointingly Flat Miss incidents

Cheap Disappointment* Frustrate Negate

Complaint Disappoints* Frustrating Negated

Complaints Disaster Frustration Negative*

Concern Disasters Grapple Negatives*

Concerned Discouragement Grappled Negligible

Concerns Discouragements Hamper Obstacle*

Counter Disliked Hampered Obstacles*

Criticism Disposal Harm Outflow

Decline* Disruption Harmed Outflows

Declined* Disturbance Harms Painful

Decliner* Doubt Harsh Penalties*

Declines* Doubts Headwind Penalty*

Declining* Down* Heavily Pessimistic

Decrease* Downgrade Hurdle* Poor

Decreased* Downgraded Hurdles* Poorly

Decreases* Downgrades Hurt Problem

Decreasing* Downside Hurted Problems

Deficit Downturn* Hurts Recession

Delay Drop* Impossible Reduce

Delayed Dropped* Inefficiencies Reduced

Delays Dropping* Inefficiency Reducing

Dent Drops* Inefficient Reduction

Dented Erode Issue Regret

Dents Eroded Issue Regret

45

Regrets Suffered Worsen*

Resilient Suffers Worsening*

Risk* Tapered Worsens*

Risks* Tapering Worst*

Risky* Terrible Worthless

Rough Terribly Wreck

Sad Threat*

Sadly Threatened

Sadness Threatens

Scandal Threats*

Scrambling Tough

Setback Tougher

Setbacks Unable

Severe Unfortunately

Shortage Unacceptable

Shrink* Uncertain*

Shrinking* Uncertainty*

Shrinks* Uncomfortable

Shrunk* Under*

Slow Underestimated

Slowdown Underestimation

Slowing Underperform

Sluggish Underperformance

Slump* Underperformed

Slumped* Undervaluated

Slumping* Undervaluation

Slumps* Unfavorable*

Smaller* Unfavourable

Smallest* Unprofitable

Soft Unsettled*

Softening Volatile

Softens Volatility

Softer Weak*

Sombre Weaken*

Squeezed Weakened*

Starve Weakening*

Starved Weakens*

Stress Weaker*

Struggle Weakness*

Struggled Weaknesses*

Subdued Worries

Suboptimal Worry

Suffer Worse*

46

Appendix 3

Figure 3.5.2.1 Scatterplot Matrix, incl. non-winsorized variables

Figure 3.5.2.2 Scatterplot Matrix, incl. winsorized variables (EARN, BTM, RET)

TONE

EARN

SIZE

BTM

RET

LOSS

ABTONE

ABACC

ABPOS

0 .5 1

-1

0

1

2

-1 0 1 2

4

6

8

4 6 8

-5

0

5

-5 0 5

0

5

10

0 5 10

0

.5

1

0 .5 1

-.5

0

.5

-.5 0 .5

0

50

100

0 50 100

0

.5

1

TONE

EARN

SIZE

BTM

RET

LOSS

ABTONE

ABACC

ABPOS

0 .5 1

-.4

-.2

0

.2

-.4 -.2 0 .2

4

6

8

4 6 8

0

2

4

0 2 4

-1

0

1

2

-1 0 1 2

0

.5

1

0 .5 1

-.5

0

.5

-.5 0 .5

0

50

100

0 50 100

0

.5

1

47

Table 4.1 Descriptive Statistics incl. non-winsorized data

Number of observations: 415

Variable Mean Median Std.Dev P1 P99 Skewness Kurtosis

TONE 0.6560 0.6761 0.2056 -0.2121 1 -0.9717 4.5327

EARN 0.0502 0.0473 0.1583 -0.3397 0.2686 5.1611 101.0054

SIZE 5.9124 5.9307 0.9034 3.2646 8.2393 0.0104 3.0090

BTM 0.6797 0.5176 0.7577 -0.1545 3.5975 1.4098 25.9760

RET 0.2984 0.2411 0.6600 -0.6777 2.0455 6.9024 83.4395

LOSS 0.1590 0 0.3662 0 1 0.1341 4.4770

ABTONE -1.06e-10 0.0109 0.1730 -0.5131 0.4799 -0.6743 4.2268

ABACC 0.9353 0.4047 4.4958 0.0002 87.5967 17.5240 334.9104

ABPOS 0.5277 1 0.4998 0 1 -0.1110 1.0123

Table 4.2 Pearson’s Correlation Analysis incl. non-winsorized data Number of observations: 415

TONE EARN SIZE BTM RET LOSS ABTONE ABACC ABPOS

TONE 1

EARN 0.2789* 1

(0.0000)

SIZE 0.3787* 0.1912* 1

(0.0000) (0.0000)

BTM -0.3384* 0.2753* -0.3660* 1

(0.0000) (0.0000) (0.0000)

RET 0.2019* 0.1661* 0.0188 -0.2170* 1

(0.0000) (0.0007) (0.7029) (0.0000)

LOSS -0.2535* -0.4493* -0.2710* -0.2937* -0.0698 1

(0.0000) (0.0000) (0.0000) (0.0000) (0.1560)

ABTONE 0.8414* 0.0464 0.0000 -0.0122 0.0002 -0.0291 1

(0.0000) (0.3458) (1.0000) (0.8038) (0.9964) (0.5540)

ABACC 0.0114 0.0301 -0.0193 0.0407 -0.0262 -0.0415 0.0400 1

(0.8165) (0.5405) (0.6945) (0.4082) (0.5943) (0.3991) (0.4158)

ABPOS 0.6242* 0.0135 -0.0542 0.0096 0.0498 0.0155 0.7694* 0.0277 1

(0.0000) (0.7839) (0.2708) (0.8452) (0.9318) (0.7536) (0.0000) (0.5734)

* Indicate significance at the 5% level

48

𝐻1: 𝑇𝑂𝑁𝐸 = 𝛽0 + 𝛽1𝐸𝐴𝑅𝑁 + 𝛽2𝑆𝐼𝑍𝐸 + 𝛽3𝐵𝑇𝑀 + 𝛽4𝑅𝐸𝑇 + 𝜀 (1)

𝐻2: 𝑃(𝐴𝐵𝑃𝑂𝑆 = 1) = 𝛽0 + 𝛽1𝐸𝐴𝑅𝑁 + 𝛽2𝐿𝑂𝑆𝑆 + 𝛽3𝑆𝐼𝑍𝐸 + 𝛽4𝐵𝑇𝑀 + 𝛽5𝑅𝐸𝑇 + 𝛽6𝐸𝐴𝑅𝑁_𝐿𝑂𝑆𝑆 + 𝜀 (2)

𝐻3: 𝐴𝐵𝑇𝑂𝑁𝐸 = 𝛽0 + 𝛽1𝐴𝐵𝐴𝐶𝐶 + 𝛽2𝐸𝐴𝑅𝑁 + 𝛽3𝐿𝑂𝑆𝑆 + 𝛽4𝑆𝐼𝑍𝐸 + 𝛽5𝐵𝑇𝑀 + 𝛽5𝑅𝐸𝑇 + 𝛽6𝐸𝐴𝑅𝑁𝐿𝑂𝑆𝑆 + 𝜀 (3)

Table 4.3 Multiple Regression Analysis incl. non-winsorized data

VARIABLES TONE (1) ABPOS (2) ABTONE (3)

EARN 0.1939* 0.3120 -0.1071*

(0.032) (0.720) (0.002)

SIZE 0.0655** -0.1031 -0.0039

(0.000) (0.413) (0.708)

BTM -0.0368** 0.0038 0.0033

(0.005) (0.981) (0.805)

RET 0.0311* 0.1717 -0.0005

(0.013) (0.346) (0.963)

LOSS 0.0881 -0.0283

(0.794) (0.424)

EARN_LOSS -0.0843 0.2770

(0.959) (0.130)

ABACC 0.0014*

(0.031)

Observations 415 415 415

R-squared 0.2900 0.0102 0.0169

* Indicate significance at the 5% level

** Indicate significance at the 1% level

Robust p-statistics in parentheses

Industry effects are included in all regressions but not reported

49

Appendix 4

Datastream Code Description

WC04860 Cash Flow from Operations

WC02999 Total Assets

WC01001 Net Sales

WC02051 Receivables

WC02501 Property, Plant & Equipment

WC08001 Market Capitalization

WC03501 Book value of Equity

WC05001 Market Price Year End

WC01751 Net Income

DPS Dividends Per Share

FTSL2C Two-digit ICB Code

ICBIC One-digit ICB Code